Research & Scientific Studies
We tackle the hardest quantitative frontiers in computational drug discovery, toxicology, and structural biology—validating our methods in blind global benchmarks, peer-reviewed preprints, and regulatory-grade QMRF dossiers.
In-Depth Research Studies & Technical Reports
Explore our complete scientific monographs detailing what we did, why we did it, our scientific methodology, benchmark evidence, and what's next.
OpenADMET Cytochrome P450 (CYP) Inhibition Blind Challenge — Global Rank #1 in Time-Dependent Inhibition, Top-10 Direct Potency & Tier 1 Cryo-EM Poses
Across 6,034 submissions (~13.8M predictions) on 6,896 compounds and 20 novel CYP3A4 Cryo-EM structures, Rasayan Labs was one of only 17 teams worldwide to make the final leaderboard across all 3 tracks using strictly public data—winning Global Rank #1 Worldwide (Tier 1) in Time-Dependent Inhibition with 100.0% CYP2D6 precision.
- •Rank #1 Worldwide in Track 2 TDI (MA-MCC 0.5887 Live · 0.4980 Final · 100% CYP2D6 Precision)
- •Rank #8 Overall (#6 Public-Data) in Direct pIC50 (CYP3A4 R² = 0.759, ρ = 0.877 · CYP2C9 R² = 0.707)
- •Tier 1 in Track 3 CYP3A4 Cryo-EM Pose Prediction (LDDT-PLI 0.4246 · Pocket LDDT-LP 0.8070)
OpenADMET PXR (NR1I2) Induction Blind Challenge — Tier 1 Finish, Statistically Tied with the Challenge Winner
Prospective blind evaluation on 513 analogue-enriched compounds (260 strictly blinded in Phase 2) combined with chemical characterization of 4,139 training records, 3D pocket complementarity, and quantum pose interaction energetics.
- •Tier 1 Finish (Ranks 9, 10 & 11 of 100) · Statistically tied with #1 (bootstrap p = 0.49)
- •Blind Phase 2 (N = 260): MAE 0.4172 · RAE 0.5780 · R² 0.558 · Spearman ρ 0.811
- •Operational Activator Screening (pEC50 ≥ 5): Retrospective ROC-AUC 0.903 · AP 0.906
ICH M7(R2)-Compliant Dual (Q)SAR System for Bacterial Reverse Mutation (RS_BMUT & RS_Expert) with Automated Impurity & Metabolite Weight-of-Evidence
Trained and calibrated on 23,009 curated OECD TG 471 compounds across S. typhimurium TA98, TA100, TA1535, TA1537 (±S9) with E. coli WP2/TA102 A–T base-pair coverage, 120M-compound read-across, CYP450 metabolite scoring, and automated process-impurity enumeration.
- •Locked 1,000-Compound Blind Hold-Out (RS_BMUT): AUC-ROC 0.9335 · Specificity 91.17% · Concordance 87.70%
- •Y-Scrambling Permutation Control: Blind AUC collapses to 0.4539 (MCC = 0.0000 · PASS)
- •RS_Expert Rule-Based Engine: 53 mechanistic alerts + 3 Cohort-of-Concern overrides + 6 dismissal rules
AI/ML-Assisted Computational Design and Immunoinformatics Evaluation of a Multi-Epitope Vaccine (RasIC-01v) Targeting Podoplanin in Glioblastoma Multiforme
Designed a 293-amino-acid subunit vaccine construct (RasIC-01v) incorporating 5 HTL, 8 CTL, and 4 B-cell epitopes with dual terminal Hp91 adjuvants targeting Podoplanin (PDPN) in Glioblastoma Multiforme, validated via glycosylated TLR3 docking and dual 100 ns explicit-water MD simulations.
- •97.0% Global HLA Coverage across 27 MHC Class I & 7 HLA-DR reference alleles (VaxiJen up to 1.4824)
- •3D Refinement in OCSR.ai ProteinLab: 98.6% Ramachandran favored (PROCHECK 95.9% core, 0.0% disallowed)
- •Glycosylated TLR3 (1ZIW) Docking (-248.55) · 100 ns MD · MM/GBSA ΔG = -20.73 kJ/mol · CAI = 0.87 in pcDNA3
NIH Tox21 Data Challenge — Ranked 5th Globally in 12-Endpoint Toxicity Screening, Open-Sourced on Hugging Face
Evaluated across 11,764 training, 296 validation, and 7,831 core compounds spanning 7 Nuclear Receptor (NR-AR, ER, AhR, Aromatase, PPAR-γ) and 5 Cellular Stress Response (SR-ARE, ATAD5, HSE, MMP, p53) pathways under 12.7:1 class imbalance.
- •Ranked 5th Globally on the NIH Tox21 Data Challenge across all 12 toxicity endpoints
- •40-Fold Stratified Cross-Validation ROC-AUC: 0.882 ± 0.021
- •Open-sourced under Apache 2.0 on Hugging Face & deployed free on OCSR.ai
AI-Designed Family of Dual COX-2/mPGES-1 Inhibitors to Treat Inflammation with Greater Precision and Effectiveness
Compressed anti-inflammatory hit discovery from 2–3 years to 3 weeks by generating 50,000 de novo structures conditioned against 2,500 off-target liabilities (COX-1, hERG, CYP2C9, CYP3A4) and docking across 8 crystal structures of human COX-2, COX-1, and trimeric mPGES-1.
- •23 High-Confidence Dual COX-2 / mPGES-1 Lead Candidates (MAI Series) with >100-fold COX-1 selectivity
- •Top Hits: MAI_2 (-10.5 kcal/mol in COX-2 5KIR, beating rofecoxib -10.1) & MAI_8 (-10.31 kcal/mol in mPGES-1 8PYV)
- •92.9% Structural Novelty (Tanimoto < 0.8) verified across 44.3M patent documents & 1.7B compounds
Darvi-Gen: 863B Flagship Model Built for Scientific Research, Autonomous Retrosynthesis & Enterprise R&D
Introducing Darvi-Gen (863.33B parameters, 256,000-token active context window, 1,696.2 tok/s cluster throughput), powering USPTO-patented lab notebook digitization (US 12,664,813 B1), RFP automation, autonomous retrosynthesis, and end-to-end drug discovery.
- •58.0% on HLE-Full w/ Tools · 91.7% GPQA-Diamond · 96.7% AIME 2026 · 83.3% SWE-Bench Verified
- •ChemBench Full 2,785-Question Corpus (Raw w/o Tools): 88.74% Core 7 Subfields · 98.31% IChO Olympiad
- •Powers USPTO Patent No. 12,664,813 B1 (30 Claims) for Chemical Intelligence & Lab Digitization
Preprints & Monographs
OpenADMET Cytochrome P450 (CYP) Inhibition Blind Challenge: Global Rank #1 in Time-Dependent Inhibition & First-Principles Thermodynamic Analysis
Rasayan Labs Research Team
Cytochrome P450 (CYP3A4, CYP2D6, CYP2C9, CYP1A2) inhibition and mechanism-based Time-Dependent Inhibition (TDI) account for over 30% of clinical drug–drug interaction failures. Across 6,034 submissions on 6,896 compounds and 20 novel CYP3A4 Cryo-EM structures, Rasayan Labs finished #1 Worldwide (Tier 1) in Time-Dependent Inhibition with 100.0% CYP2D6 precision, #8 Overall (#6 public-data) in pan-isoform direct potency, and Tier 1 in Cryo-EM ligand pose prediction.
Key Contributions
- Global Rank #1 Worldwide (Tier 1) in Track 2 TDI (MA-MCC 0.5887 live / 0.4980 final, 100.0% CYP2D6 precision)
- Proved pH 7.4 Boltzmann cationic enrichment (11.55× in CYP2D6, p = 1.54 × 10⁻³²) vs. log D7.4 drive in CYP3A4/2C9
- Resolved 73.31% single-isoform DRC sparsity & 15.9× floor censoring noise across 6,146 training compounds
- Rank #8 Overall (#6 Public-Data) in Direct Potency (CYP3A4 R² = 0.759, ρ = 0.877) & Tier 1 in Track 3 Cryo-EM Poses
News & Benchmark Achievements
OpenADMET CYP450 Blind Challenge — Global Rank #1 Worldwide in Time-Dependent Inhibition & Top-10 Direct Potency
By Rasayan Labs
In the official wrap-up of the OpenADMET Cytochrome P450 Inhibition Blind Challenge (6,034 submissions, ~13.8M predictions across 6,896 compounds and 20 novel CYP3A4 Cryo-EM structures), rasayan-labs was one of only 17 teams worldwide to place on the final leaderboard across all 3 tracks. Using strictly public data, our first-principles thermodynamic and assay-physics framework won Global Rank #1 Worldwide (Tier 1) in Track 2 Time-Dependent Inhibition (achieving 100.0% precision on CYP2D6 with zero false positives), Rank #8 Overall (#6 Public-Data) in Track 1 Direct Potency, and Tier 1 in Track 3 Cryo-EM Structure Prediction.
Key Highlights
- Global Rank #1 Worldwide (Tier 1) in Track 2 Time-Dependent Inhibition (MA-MCC 0.5887 live / 0.4980 final)
- 100.0% blind positive predictive value (zero false positives) and 95.60% accuracy on CYP2D6 TDI
- Rank #8 Overall (#6 Public-Data) in Direct pIC50 (CYP3A4 R² = 0.759, Spearman ρ = 0.877; CYP1A2 Rank #6)
- Tier 1 finish in Track 3 blind CYP3A4 Cryo-EM ligand pose prediction (LDDT-PLI 0.4246, LDDT-LP 0.8070)
Stay Updated with Our Research
Follow our journey as we push the boundaries of AI-driven drug discovery and share our findings with the scientific community.