A non-profit research initiative decoding biochemical processes through AI. All work is open-source and freely available — science should benefit everyone, not shareholders.
All our research, code, datasets, and publications are freely available under open licenses. We believe that AI-driven medical research must remain a public good — not locked behind paywalls or corporate interests. No investors. No shareholders. No profit motive. Just science for the benefit of humanity.
Six interconnected research streams. Each one a node in a larger network of discovery.
Deep learning predicts protein conformations. We analyze binding pockets to identify drug targets for cancer, autoimmune disease, and neurodegeneration.
Graph neural networks model metabolic networks. We find dysregulation patterns and intervention points invisible to traditional approaches.
Attention-based architectures predict molecular binding affinities. Screening millions of compounds — collapsing discovery from years to weeks.
Large language models on genomic sequences. Identifying clinically relevant variants for rare diseases and oncology — precision medicine at sequence level.
Reinforcement learning for adaptive trial design. Dynamic allocation, dosing, endpoints — improving power while cutting time-to-approval.
Biochemical markers + medical imaging in unified AI pipelines. Histopathology, radiology, molecular data fused for superior diagnostic accuracy.
Hypothesis → Computation → Wet Lab → Clinic. Structured. Reproducible.
Identify molecular targets via literature mining, knowledge graphs, and domain experts.
PDB, UniProt, TCGA, GEO, ChEMBL + clinical partnerships. Quality-controlled and harmonized.
Purpose-built ML architectures for the biochemical domain. Not off-the-shelf.
Cross-validation, ablation, SOTA comparison. We don't publish unless we beat the baseline.
Biochemical assays, cell experiments, animal model validation of computational predictions.
Diagnostic tools, treatment protocols, trial designs. The final mile from bench to bedside.
Infrastructure for molecular modeling and large-scale deep learning.
From working papers to peer review.
Novel GNN with 3D geometry + pocket features. 15% over DeepDTA on Davis & KIBA.
Metabolomics + transcriptomics + proteomics via graph transformers. Early dysregulation markers.
Fine-tuned genomic LLMs — pathogenic variant classification beyond ACMG.
RL-based dose-finding validated against 3+3 designs in extensive simulation.
This is a non-profit, open-source initiative. We seek collaborators from biochemistry, medicine, and computational biology who share our belief that medical AI research should remain free and accessible.