Transporters

93.6% overall accuracy across 2,376 validated drug-transporter pairs spanning 56 transporters and 8 barrier systems.

A Different Approach to an Old Problem

Most computational pharmacokinetic tools treat a transporter as a single entity — if a drug is a BCRP substrate, that label is applied uniformly regardless of where in the body BCRP is acting. The published literature tells a more complicated story. BCRP is positioned in the apical membrane of intestinal epithelial cells, where it limits oral absorption of its substrates, and is simultaneously expressed in the mammary gland during lactation, where it actively concentrates those same substrates into breast milk. ScienceDirect microenvirmonments into account to produce never befor seen results.

A Different Approach to an Old Problem

Transporters by Barrier System

Placental Transporter Screening — Per-Transporter Validation Detail

P-gp - 84.8%

Drugs Tested 116
Accuracy 94.8%
True positives 66
True negatives 44
False positives 3
False negatives 3

Three false negatives were observed: atorvastatin, phenytoin, and prednisone. Three false positives were observed: diazepam, lidocaine, and piroxicam. The false negative pattern is consistent with known literature challenges in predicting P-gp substrates among corticosteroids and select lipophilic compounds. This limitation is identified as a priority for the next development cycle.

BCRP - 96.1%

Drugs Tested 51
Accuracy 96.1%
True positives 32
True negatives 17
False positives 2
False negatives 0

Zero false negatives. Two false positives were observed: diltiazem and ibuprofen. No known BCRP substrates were missed in this validation set.

MRP2 - 93.9%

Drugs Tested 33
Accuracy 93.9%
True positives 16
True negatives 15
False positives 1
False negatives 1

One false positive: diltiazem. One false negative: pravastatin, a hydrophilic statin with limited transporter-mediated efflux in the literature — a known edge case for this transporter class.

OCT3 - 97.1%

Drugs Tested 35
Accuracy 97.1%
True positives 16
True negatives 18
False positives 1
False negatives 0

Zero false negatives. One false positive: phenytoin, a weak base with borderline cationic character. Predictions account for trimester-dependent OCT3 expression, which increases approximately two-fold from first trimester to term. Worst-case trimester exposure is used for classification.

ENT 1/2 - 100%

Drugs Tested 67
Accuracy 100%
True positives 12
True negatives 55
False positives 0
False negatives 0

Perfect accuracy across the full validation set. The ENT substrate profile is highly selective and well-characterized in the literature, producing a clean classification boundary.

Known System Limitations

Trimester-dependent transporter expression. P-gp expression decreases approximately 55% and BCRP approximately 69% between first trimester and term, based on published human placental data. This means fetal drug exposure via these protective efflux transporters is substantially higher in late pregnancy than first trimester predictions alone would suggest. The system uses trimester-aware scoring where expression data are available and flags this distinction in clinical output.

Passive diffusion. Small, lipophilic, and uncharged drug molecules can cross the placental barrier independent of transporter activity. The system applies biological override rules for known passive-diffusion drug classes — including corticosteroids, small fatty acids, and small neutral compounds — but passive transfer remains a contributing factor that transporter predictions alone do not fully capture.

Corticosteroid P-gp classification. Corticosteroids as a class present a recognized challenge for P-gp substrate prediction, consistent with findings reported in the published literature. Prednisone was the identified false negative in this validation set. Expanded corticosteroid-specific pattern coverage is the current top development priority.

Breast Milk Transporter Screening — Per-Transporter Validation Detail Version 2.2 · March 20, 2026 · Combined accuracy: 356/381 (93.4%) across all 7 transporters

BCRP - 89.3%

Drugs Tested 56
Accuracy 89.3%
True positives 27
True negatives 23
False positives 3
False negatives 3

BCRP is the primary drug-concentrating transporter in the mammary epithelium. Three false positives were observed: delavirdine, elvitegravir, and dolutegravir. These compounds share a structural binding affinity for BCRP but are documented in the published literature as non-transportable across mammary epithelial tissue — a known distinction between BCRP binding and BCRP-mediated transport that is an identified refinement target for the next development cycle. Three false negatives were observed: topotecan, acyclovir, and sulfamethoxazole, each representing borderline or weak BCRP affinity profiles consistent with their characterization in the literature.

MRP 2 - 93.9%

Drugs Tested 49
Accuracy 93.9%
True positives 23
True negatives 23
False positives 1
False negatives 2

Strong performance in the curated Phase A milk-specific validation set. One false positive was observed: digoxin, a cardiac glycoside whose steroid-like structural scaffold triggers MRP2 recognition despite not being a true milk MRP2 substrate. Two false negatives were observed: cisplatin, a platinum coordination compound whose inorganic chemistry falls outside the scope of organic-based substrate recognition — a system-wide limitation described below — and sorafenib glucuronide, a near-threshold borderline case. The lower Phase B cross-validation result of 81.8% reflects label noise in the general MRP2 dataset, which includes substrates active in liver and kidney but not in mammary tissue. Specificity in Phase B remains 100% — zero false positives — confirming the Phase A performance is not inflated.

OCT 1 - 93.8%

Drugs Tested 48
Accuracy 93.8%
True positives 25
True negatives 20
False positives 3
False negatives 0

Zero false negatives — the system never missed a true OCT1 substrate across the entire Phase A validation set. All three errors were false positives: carbamazepine, rifampicin, and phenytoin. Each carries nitrogen-containing heterocyclic structures that trigger cation transporter recognition despite not being true OCT1 substrates. This is an identified pattern shared with MATE1 and is a current development priority. The 100% sensitivity result means the system is correctly protective — it never under-calls cationic drug uptake into milk.

CNT 1 - 92.0%

Drugs Tested 50
Accuracy 92.0%
True positives 21
True negatives 25
False positives 0
False negatives 4

Zero false positives — the system never incorrectly flagged a non-substrate as a CNT1 substrate. All four errors were false negatives: adenosine, tegafur, emtricitabine, and gemcitabine. Adenosine is a purine nucleoside and CNT1 is selectively pyrimidine-preferring, making this a biologically expected boundary case rather than a model failure. Tegafur is a prodrug with a masked nucleoside structure, and emtricitabine and gemcitabine are structurally modified nucleosides where halogenation and dideoxy modifications reduce substrate recognition — a pattern shared across nucleoside prodrugs and analogs system-wide. The 100% specificity means CNT1 predictions carry no false alarm risk.

CNT 3 - 98.0%

Drugs Tested 50
Accuracy 98.0%
True positives 24
True negatives 25
False positives 0
False negatives 1

The highest-performing transporter in the validation set. Zero false positives across Phase A. One false negative: a single borderline compound. Phase B independent cross-validation confirmed the result at 95.9%, with 100% specificity — no false positives in either validation phase. CNT3 accepts both purine and pyrimidine nucleoside analogs and is clinically significant for antiviral and chemotherapy agents used during lactation. Three prodrug-related false negatives appeared in Phase B cross-validation — ribavirin, sofosbuvir, and valacyclovir — consistent with the nucleoside prodrug recognition limitation described below.

MATE 1 - 94.8%

Drugs Tested 116
Accuracy 94.8%
True positives 66
True negatives 44
False positives 3
False negatives 3

P-gp - 94.8%

Drugs Tested 116
Accuracy 94.8%
True positives 66
True negatives 44
False positives 3
False negatives 3

Blood-Brain Barrier Transporter Screening — Per-Transporter Validation Detail March 2026 · Overall accuracy: 294/313 (94.2%) across 8 active transporters

LAT 1 - 90.5%

Drugs Tested 21
Accuracy 90.5%
True positives 8
True negatives 11
False positives 0
False negatives 2

LAT1 is the primary carrier-mediated influx transporter at the blood-brain barrier for large neutral amino acids and amino acid-mimicking drugs, and is clinically significant for CNS drug delivery and levodopa transport. Zero false positives — the system never incorrectly flagged a non-substrate. Two false negatives were observed: apomorphine and ketamine, both structurally atypical LAT1 substrates whose physicochemical profiles fall at the boundary of standard amino acid mimicry recognition. The 100% specificity means LAT1 predictions carry no false alarm risk for non-substrates. The validation set of 17 drugs is noted as small and confidence intervals are wider than for larger sets — expanded LAT1 validation is an active development priority.

MCT1 - 96.4%

Drugs Tested 28
Accuracy 96.4%
True positives 17
True negatives 10
False positives 0
False negatives 1

MCT1 mediates the transport of lactate, ketone bodies, and monocarboxylate-containing drugs across the blood-brain barrier and is clinically relevant for energy substrate delivery to the CNS and for drugs including valproic acid and certain beta-hydroxy acids. Zero false positives across the full validation set. One false negative was observed: pravastatin, a hydrophilic statin whose monocarboxylate character is partially masked by its overall molecular profile. The 100% specificity and 94.4% sensitivity represent a well-balanced and highly accurate classifier for this transporter.

OATP 2B1 - 93.9%

Drugs Tested 33
Accuracy 93.9%
True positives 19
True negatives 12
False positives 1
False negatives 1

OATP2B1 at the blood-brain barrier contributes to the CNS uptake of organic anions including statins and several drug conjugates. One false positive was observed: atenolol, a beta-blocker with structural features that partially overlap the OATP2B1 substrate profile despite not being a confirmed CNS transporter substrate. One false negative was observed: asunaprevir, an HCV protease inhibitor whose complex structural scaffold places it at the edge of OATP2B1 recognition. The balanced sensitivity and specificity reflect a reliable classifier across the validated drug space.

GLUT 1 - 95.7%

Drugs Tested 23
Accuracy 95.7%
True positives 13
True negatives 9
False positives 1
False negatives 0

GLUT1 is the dominant glucose transporter at the blood-brain barrier and is responsible for the majority of CNS glucose supply. It also transports several glucose-mimicking drugs and is relevant to GLUT1 deficiency syndrome pharmacology. Zero false negatives — every confirmed GLUT1 substrate in the validation set was correctly identified. One false positive was observed: L-glucose, which is correctly not transported by GLUT1 in vivo because GLUT1 is stereospecific for D-glucose. This is a known limitation — stereoisomer pairs that differ only in spatial configuration represent a defined edge case. For all standard D-configuration glucose analogs and mimetics the system performs at 100% sensitivity.

P-gp - 95.0%

Drugs Tested 80
Accuracy 95.0%
True positives 51
True negatives 25
False positives 3
False negatives 1

P-gp is the most clinically significant efflux transporter at the blood-brain barrier and a primary determinant of CNS drug penetration for a broad range of substrates including oncology agents, HIV antiretrovirals, cardiac glycosides, and immunosuppressants. Three false positives were observed: lidocaine topical, phenytoin, and piroxicam — each sharing structural features with known P-gp substrates. One false negative was observed: cyclosporine, a large cyclic peptide whose molecular architecture falls outside the scope of standard small-molecule prediction — a known limitation where the published literature unambiguously confirms P-gp substrate status and clinical users should apply that knowledge directly. The 98.1% sensitivity is the standout result — the system correctly identified 51 of 52 true P-gp substrates in the validation set.

BCRP - 94.9%

Drugs Tested 39
Accuracy 94.9%
True positives 27
True negatives 10
False positives 0
False negatives 2

BCRP is co-expressed with P-gp at the luminal membrane of brain capillary endothelial cells and acts synergistically with P-gp to limit CNS penetration of a broad range of substrates. Zero false positives across the full validation set and 100% specificity. Two false negatives were observed: acyclovir and sulfamethoxazole, both borderline BCRP substrates with weak structural affinity profiles consistent with their characterization in the published literature. The BCRP module was recently reactivated and the 94.9% result confirms full validation.

MRP 2 - 89.1%

Drugs Tested 4
Accuracy 89.1%
True positives 30
True negatives 11
False positives 1
False negatives 4

MRP2 at the blood-brain barrier mediates efflux of organic anion conjugates and contributes to the protection of CNS tissue from accumulation of drug metabolites and xenobiotics. One false positive was observed: fentanyl, a highly lipophilic opioid whose structural profile partially overlaps MRP2 recognition despite passive diffusion being its primary BBB crossing mechanism. Four false negatives were observed: baicalin, furosemide, lamivudine, and tenofovir. The nucleoside analog false negatives — lamivudine and tenofovir — reflect a known limitation with small nucleoside analogs that is consistent across multiple modules and is an active development priority.

MRP 4 - 96.3%

Drugs Tested 57
Accuracy 96.3%
True positives 26
True negatives 29
False positives 1
False negatives 1

MRP4 is expressed at the luminal membrane of brain capillary endothelial cells and plays a particularly important role in the efflux of cyclic nucleotides, prostaglandins, and nucleoside analog antivirals from the CNS — making it clinically relevant for antiviral and antineoplastic agents used in CNS indications. One false positive was observed: amoxicillin, a beta-lactam antibiotic whose structural features partially overlap the MRP4 substrate profile. One false negative was observed: azathioprine, an immunosuppressant prodrug whose active metabolite rather than the parent compound is the primary MRP4 substrate — a known limitation of parent-compound-only prediction. The balanced 96.3% sensitivity and 96.4% specificity represent the most symmetrically accurate result in the BBB module.

Known System Limitations

Large cyclic peptides and polymer-class drugs. Drugs with unusually large or polymeric molecular architectures — including cyclosporine — fall outside the scope of standard small-molecule prediction. These compounds are known P-gp substrates per the published literature and clinical users should apply that knowledge directly.

MRP2 cross-species validation. Human-specific MRP2 experimental data is limited and the validation reflects available published substrate labels. Accuracy is expected to improve as additional human-specific data becomes available.

Small validation sets for LAT1 and GLUT1. LAT1 was validated on 17 drugs and GLUT1 on 23 drugs. Accuracy estimates for these two transporters carry wider confidence intervals than the larger sets. Expanded validation for both is underway.

Stereoisomer discrimination. Stereoisomer pairs that differ only in spatial configuration represent a known edge case. This affects one compound in the current validation set.

Small nucleoside analog recognition. Small nucleoside analogs including lamivudine and tenofovir present a consistent challenge across multiple transporter modules. This is a cross-system pattern and an active development priority.

Parked transporters. OATP1A2, OAT3, CNT2, ENT1, and MRP1 are implemented in the BBB module but are not currently validated due to insufficient labeled data. These transporters will be activated as validation data becomes available.

Hepatic Transporter Screening — Per-Transporter Validation Detail March 2026 · Overall accuracy: 295/324 (91.0%) across 9 active transporters · Patent pending

OATP 1B1 - 90.9%

Drugs Tested 22
Accuracy 90.9%
True positives 9
True negatives 11
False positives 1
False negatives 1

OATP1B1 is one of the most clinically important hepatic uptake transporters, responsible for the sinusoidal uptake of statins, sartans, and a broad range of organic anions from portal blood into hepatocytes. It is a primary determinant of hepatic first-pass extraction and a well-established site of drug-drug interactions. One false positive was observed: amoxicillin, a beta-lactam antibiotic whose structural features partially overlap the OATP1B1 substrate profile. One false negative was observed: rifampicin, a complex macrocyclic antibiotic whose large and structurally unusual scaffold places it at the edge of standard OATP1B1 recognition. The balanced sensitivity and specificity reflect a well-calibrated classifier for this transporter.

OATP 1B3 - 87.0%

Drugs Tested 23
Accuracy 87.0%
True positives 8
True negatives 12
False positives 1
False negatives 2

OATP1B3 is co-expressed with OATP1B1 at the sinusoidal membrane of hepatocytes and shares significant substrate overlap, including statins, bile acids, and several peptide-based drugs. It is independently relevant to hepatic drug clearance and interaction risk. One false positive was observed: amoxicillin, consistent with the OATP1B1 finding and reflecting a shared structural pattern. Two false negatives were observed: pravastatin and rifampicin — both representing structurally challenging compounds at the boundary of OATP1B3 recognition. Expanded validation for this transporter is an active development priority.

OATP 2B1 - 91.7%

Drugs Tested 12
Accuracy 91.7%
True positives 4
True negatives 7
False positives 0
False negatives 1

OATP2B1 contributes to hepatic uptake of organic anions and is expressed at both the hepatic sinusoidal membrane and the intestinal apical membrane. Zero false positives across the validation set. One false negative was observed: montelukast, a leukotriene receptor antagonist whose structural profile falls below standard OATP2B1 recognition thresholds. The validation set of 12 drugs is noted as small and confidence intervals are wider than for larger sets — expanded OATP2B1 validation is an active development priority.

OCT 1 - 97.0%

Drugs Tested 33
Accuracy 97.0%
True positives 23
True negatives 9
False positives 1
False negatives 0

OCT1 is the primary hepatic uptake transporter for organic cations and is clinically significant for metformin, morphine, and a range of cationic drugs whose hepatic clearance depends on sinusoidal uptake. Zero false negatives — every confirmed OCT1 substrate in the validation set was correctly identified. One false positive was observed: phenytoin, an anticonvulsant whose nitrogen-containing structure partially triggers cationic recognition despite not being a confirmed OCT1 substrate. The 100% sensitivity means the system never misses a true OCT1 hepatic substrate.

P-gp - 94.7%

Drugs Tested 75
Accuracy 94.7%
True positives 48
True negatives 23
False positives 2
False negatives 2

P-gp at the hepatic canalicular membrane mediates biliary excretion of a broad range of substrates including oncology agents, HIV antiretrovirals, and immunosuppressants — making it a key determinant of hepatic drug elimination and enterohepatic cycling. Two false positives were observed: lidocaine topical and piroxicam, consistent with the cross-system pattern of these compounds partially overlapping P-gp structural recognition. Two false negatives were observed: cisplatin and lorlatinib. Cisplatin is a platinum coordination compound representing a known limitation with inorganic chemistry described below. Lorlatinib is a macrocyclic kinase inhibitor whose constrained ring architecture places it at the edge of standard P-gp recognition. The 96.0% sensitivity confirms the system reliably identifies hepatic P-gp substrates.

BCRP - 97.2%

Drugs Tested 36
Accuracy 97.2%
True positives 26
True negatives 9
False positives 1
False negatives 0

BCRP at the hepatic canalicular membrane contributes to biliary elimination of sulfate conjugates, statins, fluoroquinolones, and several targeted oncology agents. Zero false negatives — every confirmed hepatic BCRP substrate in the validation set was correctly identified. One false positive was observed: ibuprofen, an NSAID whose structural features partially overlap BCRP recognition despite not being a confirmed hepatic BCRP substrate. The 100% sensitivity is the standout result for this transporter — no true BCRP substrate was missed across the full validation set.

MRP 2 - 97.0%

Drugs Tested 33
Accuracy 97.0%
True positives 16
True negatives 16
False positives 0
False negatives 1

MRP2 is the primary hepatic canalicular transporter for glucuronide and sulfate conjugates and organic anions, and is the transporter whose loss of function causes Dubin-Johnson syndrome. Zero false positives across the full validation set — the system never incorrectly flagged a non-substrate. One false negative was observed: valproic acid, whose acyl-glucuronide conjugate is the primary MRP2 substrate rather than the parent compound — a known limitation of parent-compound-only prediction. The 100% specificity means MRP2 predictions carry no false alarm risk for non-substrates.

BSEP - 90.9%

Drugs Tested 33
Accuracy 90.9%
True positives 2
True negatives 28
False positives 3
False negatives 0

BSEP is the primary canalicular bile salt export pump and the most clinically significant transporter for drug-induced cholestasis risk. Inhibition of BSEP — rather than substrate status — is the primary clinical concern, and BSEP inhibition screening is a separate analysis. Zero false negatives — both confirmed BSEP substrates in the validation set were correctly identified. Three false positives were observed: amoxicillin, fluvastatin, and indomethacin. It is noted that the validation set contains only two confirmed BSEP substrates, reflecting the highly selective nature of this transporter for bile acids. Sensitivity estimates carry wide confidence intervals at this sample size and expanded validation is underway.

MRP 4 - 77.2%

Drugs Tested 57
Accuracy 77.2%
True positives 18
True negatives 26
False positives 3
False negatives 10

Known System Limitations

Platinum coordination compounds. Cisplatin and other platinum-based drugs have inorganic coordination chemistry that falls outside standard organic substrate recognition. Clinical users should consult the published literature for platinum drug transporter interactions directly.

MRP4 substrate diversity. MRP4 accepts an unusually diverse range of substrates including nucleoside analogs, prostaglandins, cyclic nucleotides, and organic anions. This structural diversity presents a recognized challenge and expanded substrate coverage is an active development priority.

Parent compound prediction. The system predicts substrate status for parent drug structures. Drugs whose primary transporter interaction is through a metabolite or conjugate rather than the parent compound — such as valproic acid via its acyl-glucuronide — represent a known limitation. Metabolite SMILES can be submitted separately for independent prediction.

Small nucleoside analog recognition. Small nucleoside analogs including acyclovir, entecavir, ganciclovir, lamivudine, tenofovir, and zidovudine present a consistent challenge across multiple transporter modules due to their minimal functional group diversity. This is a cross-system pattern and an active development priority.

Small validation sets. OATP2B1 was validated on 12 drugs and BSEP on 33 drugs with only 2 confirmed substrates. Accuracy estimates for these transporters carry wider confidence intervals. Expanded validation for both is underway.

Hepatic transporter directionality. Sinusoidal transporters mediate uptake from portal blood into hepatocytes. Canalicular transporters mediate efflux into bile. MRP4 is basolateral, returning drugs to blood. The clinical consequence of substrate status differs by membrane location and direction, and Spirographic AI reports directionality explicitly for each transporter in all predictions.

Renal Transporter Screening — Per-Transporter Validation Detail March 2026 · Overall accuracy: 226/260 (86.9%) across 7 active transporters · Patent pending

OCT 2 - 86.8%

Drugs Tested 38
Accuracy 86.8%
True positives 27
True negatives 6
False positives 3
False negatives 2

OCT2 is the primary basolateral uptake transporter for organic cations at the renal proximal tubule and is the rate-limiting step for the renal secretion of many cationic drugs including metformin, cisplatin, and several antiretrovirals. It is a well-established site of clinically significant drug-drug interactions. Three false positives were observed: dolutegravir, famotidine, and quinine — each sharing cationic structural features with known OCT2 substrates. Two false negatives were observed: berberine and tenofovir. Berberine is a quaternary ammonium compound whose unusual charge distribution places it at the edge of standard OCT2 recognition. Tenofovir is a nucleotide analog whose classification at multiple renal transporters reflects the structural complexity of this drug class. The 93.1% sensitivity confirms the system reliably identifies true OCT2 substrates.

OCT 1 - 77.8%

Drugs Tested 36
Accuracy 77.8%
True positives 15
True negatives 13
False positives 3
False negatives 5

OAT1 mediates basolateral uptake of small hydrophilic organic anions from blood into proximal tubule cells and is a primary determinant of renal secretion for drugs including methotrexate, NSAIDs, and several antiviral agents. OAT1 has the lowest accuracy in the renal module at 77.8%, reflecting the structural diversity of its substrate profile and its overlap with OAT3. Three false positives were observed: cefadroxil, pravastatin, and valsartan — each sharing anionic features with known OAT1 substrates despite not being confirmed renal OAT1 transportees. Five false negatives were observed: abacavir, acyclovir, cimetidine, ganciclovir, and tenofovir. The nucleoside analog false negatives — abacavir, acyclovir, and ganciclovir — are consistent with the cross-system small nucleoside analog recognition limitation described below. Expanded OAT1 substrate coverage is a current development priority.

OCT 3 -80.6%

Drugs Tested 36
Accuracy 80.6%
True positives 19
True negatives 10
False positives 4
False negatives 3

OAT3 is co-expressed with OAT1 at the basolateral membrane of the proximal tubule and handles a partially overlapping but distinct substrate profile, preferring larger and more lipophilic anionic compounds. This structural overlap between OAT1 and OAT3 substrates contributes to cross-classification errors in both directions. Four false positives were observed: atorvastatin, cefaclor, heparin, and tenofovir. Three false negatives were observed: cimetidine, irbesartan, and torsemide — each representing compounds at the boundary between OAT1 and OAT3 selectivity or with structural profiles that fall below standard OAT3 recognition thresholds. Improved OAT1/OAT3 discrimination is an active development priority.

MATE 1 - 90.0%

Drugs Tested 30
Accuracy 90.0%
True positives 17
True negatives 10
False positives 2
False negatives 1

MATE1 mediates the final apical efflux step for organic cations from proximal tubule cells into urine, working in concert with OCT2 at the basolateral membrane to complete the tubular secretion pathway for cationic drugs. Two false positives were observed: dolutegravir and ranitidine — consistent with the cross-system pattern of nitrogen-containing compounds partially triggering cation transporter recognition. One false negative was observed: berberine, consistent with its OCT2 false negative finding and reflecting the atypical charge distribution of this quaternary ammonium compound across cation transporter modules. The 94.4% sensitivity confirms the system reliably identifies drugs that will undergo MATE1-mediated urinary excretion.

MATE 2K - 83.3%

Drugs Tested 30
Accuracy 83.3%
True positives 17
True negatives 8
False positives 2
False negatives 3

MATE2K is the kidney-specific isoform of the MATE family, expressed exclusively at the apical membrane of renal proximal tubule cells where it complements MATE1 in the urinary excretion of organic cations. Two false positives were observed: dolutegravir and ranitidine, consistent with the MATE1 false positive pattern. Three false negatives were observed: acyclovir, berberine, and ganciclovir — the nucleoside analog false negatives consistent with the cross-system pattern, and berberine consistent with the quaternary ammonium recognition limitation. The validation set of 30 drugs is noted as relatively small and confidence intervals are wider than for larger sets. Expanded MATE2K validation is an active development priority.

MRP 2 - 97.0%

Drugs Tested 33
Accuracy 97.0%
True positives 16
True negatives 16
False positives 0
False negatives 1

MRP2 at the renal apical membrane mediates urinary excretion of glucuronide and sulfate conjugates, organic anions, and several anionic drugs — complementing its role at the hepatic canalicular membrane in eliminating conjugated metabolites from the body. Zero false positives across the full validation set — the system never incorrectly flagged a non-substrate, and 100% specificity means MRP2 predictions carry no false alarm risk for non-substrates. One false negative was observed: sulfamethoxazole, a sulfonamide antibiotic whose renal MRP2 substrate status reflects a structural profile at the boundary of current recognition. The 97.0% accuracy is the strongest result in the renal module.

MRP 4 - 91.2%

Drugs Tested 57
Accuracy 91.2%
True positives 25
True negatives 27
False positives 2
False negatives 3

MRP4 at the renal apical membrane mediates urinary excretion of nucleotides, prostaglandins, cyclic nucleotides, and several organic anions — making it clinically relevant for antiviral nucleoside analogs and drugs affecting renal prostaglandin handling. Two false positives were observed: rifampicin and rosuvastatin, each sharing structural features with known MRP4 anionic substrates. Three false negatives were observed: sildenafil, torsemide, and valproic acid — each representing structural diversity at the boundary of MRP4 recognition. The balanced sensitivity and specificity at 89.3% and 93.1% respectively reflect a well-calibrated classifier for this transporter at the renal barrier.

Known System Limitations

OAT1 and OAT3 substrate overlap. OAT1 and OAT3 share a partially overlapping substrate space with distinct but not fully separable selectivity profiles. This structural overlap contributes to cross-classification errors for both transporters, particularly for drugs in the borderline anionic molecular weight range. Improved OAT1/OAT3 discrimination is an active development priority.

Small nucleoside analog recognition. Small nucleoside and nucleotide analogs including acyclovir, abacavir, ganciclovir, and tenofovir present a consistent challenge across OAT1, MATE2K, and other modules due to their minimal distinguishing functional group profiles. This is a cross-system pattern shared across multiple barrier modules and is an active development priority.

Quaternary ammonium compounds. Berberine and structurally similar quaternary ammonium compounds have atypical charge distributions that fall outside standard organic cation recognition patterns for OCT2, MATE1, and MATE2K. This is an identified structural class requiring expanded pattern coverage.

Platinum coordination compounds. Cisplatin is a platinum coordination compound whose inorganic chemistry falls outside standard organic substrate recognition. Cisplatin nephrotoxicity is well established in the published literature as OCT2-mediated and clinical users should apply that knowledge directly.

Proximal tubule scope. All seven validated transporters are located in the renal proximal tubule. Distal tubule and collecting duct processes — including passive reabsorption driven by urinary pH gradients and urine concentration — are not modeled. For weak bases, actual urinary excretion may be higher than transporter-based prediction alone would suggest under acidic urine conditions.

P-gp and BCRP at the kidney. P-gp and BCRP are expressed at the renal apical membrane and are included in pathway-level predictions but have not been individually validated at the renal barrier due to insufficient labeled data at this site. Their per-transporter accuracy is not reported and will be added as validation data becomes available.

Small validation sets for MATE1 and MATE2K. Both MATE transporters were validated on 30 drugs each. Accuracy estimates carry wider confidence intervals than the larger sets and expanded validation is underway.

Gastrointestinal Transporter Screening — Per-Transporter Validation Detail

PEPT 1 - 90.0%

Drugs Tested 50
Accuracy 90.0%
True positives 21
True negatives 24
False positives 3
False negatives 2

PEPT1 is the primary intestinal uptake transporter for dipeptides, tripeptides, and a broad range of peptidomimetic drugs including beta-lactam antibiotics, ACE inhibitors, and antiviral prodrugs. Three false positives were observed: atenolol, furosemide, and methotrexate — each sharing structural features with known PEPT1 substrates despite not being confirmed transportees. Two false negatives were observed: fosinopril and oseltamivir, both prodrugs with structural modifications that reduce standard substrate recognition. The 91.3% sensitivity confirms the system reliably identifies true PEPT1 substrates, which is the clinically important direction for oral bioavailability prediction.

OATP 2B1 - 84.9%

Drugs Tested 53
Accuracy 84.9%
True positives 20
True negatives 25
False positives 2
False negatives 6

OATP2B1 mediates intestinal uptake of a range of organic anions including statins, sartans, and several antibiotics and is relevant to food-drug interactions — notably with grapefruit juice and apple juice, which are known OATP2B1 inhibitors. Two false positives were observed: dabigatran and penicillin V. Six false negatives were observed: aliskiren, ambrisentan, bromosulfophthalein, celiprolol, sorafenib, and sulfasalazine — each representing structurally diverse compounds at the boundary of OATP2B1 substrate recognition. The 92.6% specificity indicates the system is conservative and does not over-call substrates, which limits false interaction flags.

OAP 1A2 - 97.9%

Drugs Tested 48
Accuracy 97.9%
True positives 37
True negatives 10
False positives 1
False negatives 0

Zero false negatives across the full validation set — the system correctly identified every confirmed OATP1A2 substrate in the dataset. One false positive was observed: lopinavir, an HIV protease inhibitor sharing structural features with known OATP1A2 substrates. OATP1A2 has a broad substrate profile including statins, fluoroquinolone antibiotics, thyroid hormones, and several oncology agents. The 100% sensitivity result means no true OATP1A2 substrate was missed in validation.

P-gp - 85.7%

Drugs Tested 56
Accuracy 85.7%
True positives 24
True negatives 24
False positives 4
False negatives 4

P-gp is the most clinically studied intestinal efflux transporter and a major determinant of oral bioavailability for a large number of drugs including oncology agents, HIV protease inhibitors, cardiac glycosides, and immunosuppressants. Four false positives were observed: amlodipine, furosemide, glimepiride, and pioglitazone. Four false negatives were observed: amitriptyline, clopidogrel, cyclosporine, and morphine — each representing structural profiles at the boundary of P-gp recognition or cases where complex ring systems limited prediction confidence. The perfectly balanced sensitivity and specificity at 85.7% reflects a well-calibrated classifier without bias toward either error direction.

BCRP - 88.5%

Drugs Tested 61
Accuracy 88.5%
True positives 30
True negatives 24
False positives 6
False negatives 1

BCRP is co-expressed with P-gp at the intestinal apical membrane and shares a significant number of substrates, including statins, fluoroquinolones, and several targeted oncology agents. One false negative was observed: uric acid, an endogenous BCRP substrate with atypical physicochemical properties relative to the drug substrate profile. Six false positives were observed, with three representing cases where docking-based prediction overrode the gate assessment — an identified refinement area for the docking upgrade logic. The 96.8% sensitivity is the standout result — the system correctly identified 30 of 31 confirmed BCRP substrates, meaning it is highly reliable for flagging drugs that will face intestinal BCRP-mediated efflux.

MRP 2 - 100%

Drugs Tested 27
Accuracy 100%
True positives 19
True negatives 8
False positives 0
False negatives 0

Perfect accuracy across the full validation set. Zero false positives and zero false negatives. MRP2 mediates intestinal efflux of glucuronide and sulfate conjugates, organic anions, and several antiretroviral and anticancer agents. The 100% result reflects the well-characterized and structurally distinctive substrate profile of intestinal MRP2. It is noted that the validation set for this transporter is smaller than others at 27 labeled drugs, and expanded validation is underway.

MCT 1 - 100%

Drugs Tested 37
Accuracy 100%
True positives 26
True negatives 11
False positives 0
False negatives 0

Perfect accuracy across the full validation set. MCT1 mediates intestinal uptake of short-chain fatty acids, ketone bodies, lactate, and a range of carboxylic acid-containing drugs including NSAIDs, valproic acid, and niacin. The highly selective substrate profile — defined by the monocarboxylate structural requirement — produces a clean classification boundary with zero errors across 37 drugs.

MCT 4 - 100%

Drugs Tested 37
Accuracy 100%
True positives 26
True negatives 11
False positives 0
False negatives 0

Perfect accuracy across the full validation set. MCT4 is the primary efflux transporter for lactate and other monocarboxylates at the basolateral enterocyte membrane, with a substrate profile that partially overlaps with MCT1 but with distinct selectivity characteristics. Zero errors across 49 drugs, including correct rejection of 26 structurally related non-substrates.

CNT 2 - 100%

Drugs Tested 50
Accuracy 100%
True positives 50
True negatives N/A
False positives 0
False negatives 0

Perfect identification of all 50 confirmed CNT2 substrates in the validation set. CNT2 is the primary intestinal transporter for purine nucleosides and nucleoside analog antivirals and chemotherapy agents, including abacavir, didanosine, ribavirin, and a broad range of antiviral and anticancer nucleoside analogs. It is noted that the current validation set contains confirmed substrates only — non-substrate negative labels for CNT2 are not yet available in the dataset, and specificity cannot be calculated until negative controls are added.

Known System Limitations

Transporter expression variation across gut segments. MRP2 and BCRP expression is highest in the proximal small intestine and decreases distally, while P-gp expression shows the opposite pattern, increasing from duodenum to colon. The system predicts substrate potential for each transporter independently and does not currently model segment-specific expression variation. Clinical output reflects substrate status at the most highly expressed segment.

Complex ring system conformer failures. A small number of drugs with bridged or complex polycyclic ring systems cannot generate reliable three-dimensional conformers for docking-based prediction. These compounds fall back to gate-only classification for the ABC transporters. This is an identified boundary condition rather than a model failure.

Docking upgrade directionality. For ABC transporters, the docking component can recover missed substrates but cannot correct gate-level false positives. If the gate incorrectly flags a non-substrate, the docking score will not override that call. This is reflected in the false positive profile for P-gp and BCRP.

Metabolite transport not modeled. The system predicts substrate status for parent drug structures. Phase II metabolites — glucuronides, sulfate conjugates, and other conjugated products — may have different transporter profiles than their parent compounds and require separate SMILES input for independent prediction.

Drug-drug interaction modeling. The system predicts substrate status for each compound independently. Competitive inhibition between co-administered drugs sharing a transporter — a clinically significant source of drug interactions — is not modeled in the substrate prediction output.

CNT2 negative label coverage. Specificity for CNT2 cannot currently be calculated due to the absence of confirmed non-substrate labels in the validation set. Negative control expansion for CNT2 is an active development priority.

OATP2B1 sensitivity. At 76.9%, OATP2B1 sensitivity is the lowest individual result in the gastrointestinal module. This reflects the structural diversity of OATP2B1 substrates and the limited published experimental data for this transporter compared to P-gp or BCRP. Expanded substrate coverage for OATP2B1 is a priority for the next development cycle.

Blood-Retinal Barrier Transporter Screening — Per-Transporter Validation Detail March 2026 · Overall accuracy: 199/236 (84.3%) across 5 active transporters · Patent pending

P-gp - 86.0%

Drugs Tested 79
Accuracy 86.0%
True positives 43
True negatives 27
False positives 2
False negatives 7

P-gp is a primary efflux transporter at the blood-retinal barrier, expressed at both the inner and outer retinal barrier layers, where it protects retinal tissue from accumulation of a broad range of substrates including oncology agents, antiretrovirals, and macrolide antibiotics. Two false positives were observed: lidocaine topical and piroxicam, consistent with the cross-system pattern of these compounds partially overlapping P-gp structural recognition across multiple barrier modules. Seven false negatives were observed: azithromycin, clarithromycin, erythromycin, lovastatin, simvastatin, lorlatinib, and cisplatin. The three macrolide antibiotics and two statins share large macrocyclic or extended surface contact structures that represent a known limitation described below. Cisplatin is a platinum coordination compound representing a separate known limitation with inorganic chemistry. Lorlatinib is a macrocyclic kinase inhibitor falling under the same macrocycle limitation.

BCRP - 89.7%

Drugs Tested 39
Accuracy 89.7%
True positives 26
True negatives 9
False positives 1
False negatives 3

BCRP is co-expressed with P-gp at the retinal barrier and contributes to the efflux of a broad range of substrates including statins, fluoroquinolones, and targeted oncology agents — limiting their retinal accumulation. One false positive was observed: ibuprofen, an NSAID whose aromatic and carboxyl structural features partially overlap BCRP substrate recognition, consistent with findings across other barrier modules. Three false negatives were observed: ceftriaxone, imatinib, and mitoxantrone — each representing structural classes at the boundary of current BCRP substrate coverage. Expanded kinase inhibitor and anthracenedione pattern coverage is an active development priority.

MRP 1 - 82.8%

Drugs Tested 29
Accuracy 82.8%
True positives 5
True negatives 19
False positives 0
False negatives 5

MRP1 is expressed at the basolateral membrane of the retinal pigment epithelium where it plays a protective role for the photoreceptors lying beneath. Zero false positives across the full validation set — the system never incorrectly flagged a non-substrate, and 100% specificity means MRP1 predictions carry no false alarm risk. Five false negatives were observed: bilirubin glucuronide, estrone 3-sulfate, grepafloxacin, ritonavir, and saquinavir. The glucuronide and sulfate conjugate false negatives reflect a known limitation in conjugate recognition described below. The HIV protease inhibitor false negatives reflect the macrocyclic structural complexity shared with the P-gp false negative pattern. Expanded conjugate and macrocycle coverage are current development priorities.

MRP 4 - 77.6%

Drugs Tested 58
Accuracy 77.6%
True positives 16
True negatives 29
False positives 1
False negatives 12

MRP4 at the retinal barrier plays a clinically distinctive role — its regulation of prostaglandin transport directly influences intraocular pressure, making it relevant to glaucoma pharmacology. It also mediates efflux of cyclic nucleotides, nucleoside analogs, and organic anions. One false positive was observed: amoxicillin, whose strongly anionic character partially overlaps MRP4 substrate recognition. Twelve false negatives were observed: ganciclovir, hydrochlorothiazide, leucovorin, mercaptopurine, methotrexate, olmesartan, sildenafil, thioguanine, topotecan, torsemide, uric acid, and valproic acid. The breadth of this false negative list reflects the unusually diverse MRP4 substrate profile — small organic anions, nucleoside analogs, thiopurines, and prostaglandin-related compounds each represent distinct structural classes that are collectively challenging to cover. The 96.7% specificity confirms the system does not over-call substrates. Expanded nucleotide, prostaglandin, and thiopurine pattern coverage is a current development priority and is expected to recover the majority of these false negatives.

MRP 5 - 80.6%

Drugs Tested 31
Accuracy 80.6%
True positives 6
True negatives 19
False positives 1
False negatives 5

MRP5 holds a clinically unique position in the retinal barrier system — it is the primary transporter responsible for cGMP efflux from photoreceptor cells, making it directly relevant to the phototransduction cascade and visual function. One false positive was observed: acyclovir, a nucleoside analog whose structural features partially trigger MRP5 recognition. Five false negatives were observed: abacavir, indomethacin, mercaptopurine, methotrexate, and thioguanine. The thiopurine false negatives — mercaptopurine and thioguanine — and the antifolate false negative — methotrexate — reflect a known gap in thiopurine and antifolate-specific pattern coverage that is an active development priority. The 95.0% specificity confirms the system is appropriately conservative in its MRP5 substrate calls.

Known System Limitations

Macrocyclic and large ring structures. Macrolide antibiotics including azithromycin, clarithromycin, and erythromycin, along with macrocyclic kinase inhibitors and large cyclic peptides, interact with P-gp through extensive molecular surface contact rather than the concentrated functional group interactions captured by current pattern coverage. This is a consistent cross-system pattern and macrocycle recognition expansion is an active development priority.

Conjugate recognition. MRP1 and MRP4 transport glucuronide and sulfate conjugates as primary substrates, but conjugate-specific pattern coverage is currently incomplete. Bilirubin glucuronide and estrone 3-sulfate are the identified false negatives in this category. Expanded conjugate recognition is a development priority.

MRP4 substrate diversity. MRP4 accepts an unusually broad range of substrates spanning prostaglandins, cyclic nucleotides, nucleoside analogs, organic anions, and thiopurines. This structural diversity is the primary driver of the lower MRP4 sensitivity across all barrier modules and is an active development priority.

Small nucleoside analog and thiopurine recognition. Small nucleoside analogs and thiopurines including mercaptopurine, thioguanine, and methotrexate present a consistent challenge across MRP4 and MRP5 due to their minimal distinguishing functional group profiles. This is a cross-system pattern shared with the hepatic, BBB, and gastrointestinal modules.

Stereoisomer discrimination. Substrate prediction operates on two-dimensional molecular descriptors. Stereoisomers that differ only in spatial configuration produce identical feature profiles and cannot currently be distinguished — a known limitation shared across all barrier modules.

All-efflux architecture. The blood-retinal barrier module models five efflux transporters only. Retinal penetration classification is determined by passive permeability characteristics in combination with efflux substrate status. No dedicated influx transporter is currently modeled for this barrier.

MRP5 cGMP specificity. MRP5 is the primary cGMP efflux transporter in photoreceptors and is clinically relevant to drugs that modulate phototransduction. The current model cannot distinguish cGMP structural analogs from other nucleotide analogs, which contributes to both the false positive and false negative profile for this transporter.

Pulmonary Transporter Screening — Per-Transporter Validation Detail March 2026 · Overall accuracy: 148/162 (91.4%) across 4 active transporters · Patent pending

P-gp - 90.7%

Drugs Tested 75
Accuracy 90.7%
True positives 48
True negatives 20
False positives 5
False negatives 2

P-gp at the pulmonary apical membrane occupies a clinically distinctive position compared to other barrier systems — rather than protecting a target tissue from drug accumulation, pulmonary P-gp pumps drugs back into the airway lumen, reducing their systemic absorption after inhalation. This directional distinction is reflected in clinical output for all pulmonary predictions. Five false positives were observed: lidocaine topical, piroxicam, pitavastatin, pravastatin, and rosuvastatin. The three statin false positives share a structural profile that overlaps P-gp recognition but are primarily BCRP substrates rather than P-gp substrates — clinical users evaluating statin predictions should cross-reference the BCRP result. Two false negatives were observed: cisplatin and lovastatin. Cisplatin is a platinum coordination compound representing the known inorganic chemistry limitation described below. Lovastatin is a large lactone-containing statin whose macrocyclic-adjacent structure places it at the boundary of standard P-gp recognition. The 96.0% sensitivity confirms the system reliably identifies true pulmonary P-gp substrates.

BCRP - 97.4%

Drugs Tested 39
Accuracy 97.4%
True positives 29
True negatives 9
False positives 1
False negatives 0

BCRP is co-expressed with P-gp at the pulmonary apical membrane and contributes to limiting the systemic absorption of inhaled drug substrates including statins, fluoroquinolones, and targeted oncology agents. Zero false negatives across the full validation set — every confirmed pulmonary BCRP substrate was correctly identified. One false positive was observed: ibuprofen, consistent with the cross-system pattern of this NSAID partially overlapping BCRP structural recognition across multiple barrier modules. The 100% sensitivity and 97.4% overall accuracy make BCRP the strongest result in the pulmonary module.

MRP 1 - 95.7%

Drugs Tested 23
Accuracy 95.7%
True positives 15
True negatives 7
False positives 0
False negatives 1

MRP1 at the pulmonary basolateral membrane is directionally opposite to P-gp and BCRP — it pumps drugs from airway epithelial cells into the bloodstream, increasing rather than reducing systemic absorption of inhaled substrates. This makes MRP1 clinically relevant for inhaled drug bioavailability, particularly for anticancer agents and GSH-conjugated metabolites. Zero false positives across the full validation set and 100% specificity. One false negative was observed: carboplatin, a platinum coordination compound representing the known inorganic chemistry limitation described below. The 95.7% accuracy is achieved from pattern analysis alone as no MRP1 crystal structure with a bound ligand is currently available for docking-based enhancement. The validation set of 23 drugs is noted as relatively small and expanded validation is underway.

MRP 5 - 80.0%

Drugs Tested 25
Accuracy 80.0%
True positives 8
True negatives 12
False positives 0
False negatives 5

MRP5 in the pulmonary system operates as an intracellular transporter regulating cyclic nucleotide levels — cAMP and cGMP — within airway epithelial cells, with relevance to bronchodilation signaling and inflammatory mediator control. Zero false positives across the full validation set and 100% specificity — the system never incorrectly flagged a non-substrate. Five false negatives were observed: abacavir, mercaptopurine, thioguanine, methotrexate, and indomethacin. The thiopurine false negatives — mercaptopurine and thioguanine — and antifolate false negative — methotrexate — are consistent with the cross-system pattern of these structural classes presenting minimal distinguishing functional group features, as seen in the retinal MRP5 validation. MRP5 has the lowest sensitivity in the pulmonary module at 61.5%, reflecting both the structural diversity of its substrate profile and the absence of a crystal structure available for docking-based sensitivity enhancement. Expanded thiopurine and antifolate pattern coverage is a current development priority.

Known System Limitations

Platinum coordination compounds. Cisplatin and carboplatin have inorganic coordination chemistry that falls outside standard organic substrate recognition. Both appear as false negatives — cisplatin for P-gp and carboplatin for MRP1. These drugs are established efflux pump substrates per the published literature and clinical users should apply that knowledge directly.

Statin P-gp and BCRP overlap. Several statins — pitavastatin, pravastatin, and rosuvastatin — are predicted as P-gp substrates but are primarily BCRP substrates. Their molecular profiles overlap P-gp recognition without being biologically transported by P-gp at the pulmonary barrier. Clinical users evaluating statin pulmonary predictions should review both the P-gp and BCRP outputs together.

MRP5 substrate diversity and sensitivity. MRP5 accepts thiopurines, antifolates, cyclic nucleotide analogs, and select NSAIDs — a structurally diverse profile with minimal shared functional group features. This is the primary driver of the lower MRP5 sensitivity and is consistent with findings across the retinal module. No crystal structure is available for docking-based enhancement. Expanded substrate coverage is an active development priority.

MRP1 crystal structure availability. No MRP1 crystal structure with a bound ligand is currently publicly available. MRP1 predictions are generated from pattern analysis alone without docking-based sensitivity enhancement. Accuracy is expected to improve when a suitable structure becomes available.

Large cyclic peptides and polymer-class drugs. Cyclosporine and similar large peptide or polymer drugs fall outside the scope of standard small-molecule prediction. These are known P-gp substrates per the published literature and clinical users should apply that knowledge directly.

Small nucleoside analog and thiopurine recognition. Small nucleoside analogs and thiopurines including abacavir, mercaptopurine, and thioguanine present a consistent challenge across MRP5 at multiple barrier modules. This is a cross-system pattern and an active development priority.

Pulmonary directionality context. At the pulmonary barrier, P-gp and BCRP efflux reduces systemic absorption of inhaled drugs by returning them to the airway lumen, while MRP1 efflux increases systemic absorption by pumping drugs toward the bloodstream. This is the reverse of the protective role these transporters play at most other barrier systems. Spirographic AI reports directionality explicitly for each transporter in all pulmonary predictions.

Inhaled drug delivery modeling scope. The pulmonary module predicts transporter substrate status only. Particle deposition dynamics, aerosol characteristics, mucociliary clearance, surfactant interactions, and blood-gas partition coefficients relevant to inhaled and volatile drug delivery are outside the current prediction scope.