Scientific Intelligence for Drug Discovery
Friday, April 17, 2026
MIT's open-weight model matches AlphaFold3 on protein-ligand complexes while running on a single GPU. For computational chemistry teams, the implications for hit-to-lead timelines are significant — and immediate.
The AI-designed PI3Kα inhibitor shows durable responses in HR+/HER2- breast cancer with mutant-selective specificity. A landmark for computationally designed oncology drugs.
The largest Series B in AI pharma history signals growing conviction that foundation models can compress timelines from target to IND.
Targeted protein degradation meets deep learning. Where the biology is ready, where the data is not, and what it means for drug programs targeting tau and alpha-synuclein.
The updated diffusion model achieves 58% top-1 success rate on PoseBusters, narrowing the gap with physics-based methods while maintaining speed advantages.
The collaboration aims to run millisecond-scale MD simulations for thousands of compounds simultaneously, a potential inflection point for free energy perturbation workflows.
64 AI tools tracked across 7 R&D categories. Benchmarks, maturity scores, and real-world adoption data.
Track who's using what across the AI drug discovery landscape. Pipeline-level adoption signals.
Trial outcomes that matter for AI-designed and AI-optimized drug candidates entering the clinic.
Boltz-2 now handles covalent ligands and glycosylated proteins out of the box. Ran it on a notoriously difficult kinase–inhibitor complex — RMSD under 1.2 Å. The pace of open-source structural biology is unreal.
New RFdiffusion checkpoint trained on 150K experimental structures + 2M AF2 predictions. Binder design success rates jumped from ~15% to ~40% in our wet-lab validation. Preprint dropping Friday.
Recursion's Phase II results for their AI-discovered STING agonist just dropped. ORR of 38% in refractory solid tumors. This is the first AI-native drug to show real clinical signal in a randomized trial. Landmark moment.
We're releasing our multi-modal foundation model for cell biology — trained on 40M cell images paired with transcriptomics. It predicts phenotypic outcomes from perturbations with 0.87 AUROC. Open weights on HuggingFace.
Join practitioners getting the signal, not the noise. One email per week covering tool launches, clinical readouts, and science gaps in AI drug discovery.
Evaluate where AI can realistically accelerate your drug discovery pipeline today — not in theory, but based on your current data and infrastructure.
Take the assessment14 tools
10 tools
11 tools
16 tools
10 tools
7 tools
12 tools
3 tools
3 tools
4 tools
1 tools