Research & News
We tackle the most challenging frontiers in computational drug discovery, advancing the field through innovative AI architectures and groundbreaking research that transforms how medicines are discovered.
Research Papers
AI/ML-Assisted Computational Design and Immunoinformatics Evaluation of a Multi-Epitope Vaccine Targeting Podoplanin in Glioblastoma Multiforme
Rasayan Labs & Institute of Chemical Technology, Mumbai
Glioblastoma Multiforme is one of the deadliest brain cancers, with median survival under 15 months and almost no durable therapy. We applied an AI/ML-assisted immunoinformatics pipeline to design a multi-epitope vaccine targeting Podoplanin (PDPN) — a transmembrane glycoprotein that drives glioblastoma invasion and metastasis. The construct was validated through 3D structural modelling, codon optimization, and in-silico cloning into a mammalian expression vector.
Key Contributions
- Targets Podoplanin, a key driver of glioblastoma tumour invasion
- B-cell, CTL, and HTL epitopes screened for antigenicity, non-allergenicity, and non-toxicity
- Adjuvant + linker assembly with verified expression in mammalian vectors
- Opens a low-cost, computational route toward immunotherapies for the deadliest brain cancer
News & Achievements
OpenADMET PXR Induction Blind Challenge — Tier 1 Finish, Statistically Tied with the Winner
By Rasayan Labs
Rasayan Labs finished in Tier 1 — the top statistical tier — of the OpenADMET PXR (NR1I2) Induction Blind Challenge, ranking 9 of 100 on the fully-blinded 260-compound test set with MAE 0.417, RAE 0.578, R² 0.56, and Spearman ρ 0.81. A head-to-head bootstrap comparison against the first-ranked model finds the difference statistically insignificant (p = 0.49), placing our model on par with the challenge winner. Our 12-voice stacked ensemble is led by a CheMeleon encoder pretrained on the assay's own high-throughput screen and fine-tuned with fused structural signal — pocket complementarity (DDP) and quantum pose energy (QVS) — a fusion that measurably sharpened predictions on the blind set. In the spirit of open science, the complete model report — including every one of 40+ rejected experiments and the activity-cliff analysis that defines the field's ceiling — is public.
Key Highlights
- Tier 1 (top statistical tier), rank 9 of 100 — statistically tied with the #1 model (head-to-head bootstrap p = 0.49)
- Blind 260-compound test set: MAE 0.417 · RAE 0.578 · R² 0.56 · Spearman ρ 0.81
- Led by a structure-fused, HTS-pretrained CheMeleon voice — pocket docking (DDP) + quantum pose scoring (QVS), a blind-verified gain
- Fully open: model report and 40+ ablation experiments published, characterizing the activity-cliff performance ceiling
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.