Issue #19 · Weekly Intelligence Briefing

Week 19 · May 5–11, 2026

The deals, papers, and market moves shaping AI drug discovery this week.

Deals
3 · $890M
Press Releases
127
Top Mover
RXRX −8%
Papers Ranked
10

Deals & Partnerships

Licensing, collaborations, and pipeline milestones announced this week.

Date Companies Type Value Summary
May 6 Insilico Medicine / Qilu Pharmaceutical Licensing $120M Exclusive Greater China rights to a preclinical USP1 inhibitor for oncology; upfront plus milestones and royalties.
May 7 Recursion / Roche-Genentech Collaboration $150M Expanded neuroscience deal using phenomics to prosecute three undisclosed CNS targets; upfront plus $2B biobucks.
May 9 Relay Therapeutics Milestone Phase 3 RLY-2608 (PI3Kα mutant-selective) enters pivotal Phase 3 in HR+/HER2− breast cancer; interim readout expected 1H 2027.

Company Press Releases

127 releases parsed this week. Top signals below.

Pipeline
May 5·Insilico Medicine
First generative-AI-designed drug hits primary endpoint; FVC stabilized at 12 weeks.
May 6·Recursion
MEK1/2 inhibitor enters efficacy study; 60 patients enrolled at 8 US sites.
May 9·Relay Therapeutics
Pivotal trial vs. fulvestrant+alpelisib; 540 patients, 1H 2027 interim.
May 8·AbCellera
First internal asset into the clinic; anti-KLK5 antibody for Netherton syndrome.
Platform
May 7·Isomorphic Labs
New ligand-aware structure prediction with 2× binding-pose accuracy on PDBbind.
May 6·Schrödinger
Adds protein-protein interface scoring and automated ADMET cascade.
May 10·XtalPi
Free-energy perturbation now available as API to global pharma partners.
Corporate
May 5·Tempus AI
Reiterates FY26 guidance; Genomics segment accelerates on DOD contracts.
May 8·Evotec
Portfolio consolidation focusing on J.POD biologics manufacturing and Bristol Myers partnership.
May 11·BenevolentAI
Catherine Moukheibir steps in; company reiterates 2026 cash runway to Q3 2027.

Media Coverage

How the financial and trade press framed the week.

STAT News May 7, 2026 Positive
The Phase 2b readout for INS018_055 is the strongest clinical validation yet of end-to-end generative chemistry, with implications well beyond IPF.
Endpoints May 7, 2026 Neutral
Genentech doubles down on phenomics in CNS but cuts the deal's initial indications from six to three — a more disciplined bet after last year's mixed readouts.
FierceBiotech May 8, 2026 Positive
RLY-2608's hyperglycemia profile remains a key differentiator vs. alpelisib as it enters a pivotal 540-patient breast-cancer trial.
Nature Biotech May 9, 2026 Neutral
An editorial perspective arguing that generative structure is not a substitute for target biology, and mapping the workflow where AI actually compresses timelines.
Bloomberg May 10, 2026 Critical
Sector multiples compress 12% on the week as investors demand clinical proof over platform narrative; RXRX, SDGR, ABSI lead the decline.
STAT News May 10, 2026 Neutral
Agency signals a risk-based framework for model documentation in regulatory submissions, with stricter requirements for generative models used in IND-enabling studies.
Endpoints May 11, 2026 Positive
Analysts point to runway extension and a focused oncology pipeline as reasons the CFO change reads as operational, not existential.

Top 10 Papers of the Week

Ranked by scientific novelty, translational relevance, and peer signal.

01
Nguyen, Patel, Kovacs et al. · Nature · May 7, 2026
Introduces POCKETFLOW, a continuous-time flow-matching model conditioned on 3D pocket geometry. Beats Pocket2Mol and DiffDock-L on binding-pose RMSD while generating synthesizable scaffolds.
Generative Chemistry Structure-based
02
Li, Okafor, Shah · Nat. Chem. Biol. · May 6, 2026
A multimodal encoder aligns E3 ligase and target binders with linker geometry, producing 14 novel PROTACs with <100 nM DC50 against BRD4 and SMARCA2.
PROTAC Generative Chemistry
03
Watson, Zhang, Baker · Science · May 8, 2026
Extension of ProteinMPNN that explicitly conditions on catalytic geometry. 40% of designed serine-hydrolase scaffolds show measurable activity without directed evolution.
Protein Design Enzyme Engineering
04
Alvarez, Chen, Rood et al. · Cell · May 5, 2026
Pre-trained on 6M Perturb-seq profiles across 41 cell types; identifies causal drivers with 3× higher hit rate than differential-expression baselines in a prospective pancreatic-cancer study.
Target ID Foundation Model
05
Yamashita, Riley, Bender · J. Chem. Inf. Model. · May 6, 2026
Reports state-of-the-art on MPO benchmarks with conformal prediction intervals. Open weights and curated dataset released with the paper.
ADMET Uncertainty
06
Perez, Huang, Topol · Nat. Med. · May 9, 2026
Retrospective validation across 14 Phase 2 oncology trials; external control arms matched on 400-dim patient embeddings preserve effect-size estimation.
Clinical AI Digital Twin
07
Corso, Stark, Jaakkola · ICLR 2026 · May 6, 2026
Addresses the main failure mode of rigid docking by sampling pocket conformations jointly with ligand pose. 18% top-1 RMSD improvement on PoseBusters.
Docking Diffusion
08
Walters, Gomes, Swanson · arXiv · May 7, 2026
Empirical analysis across 12 property benchmarks showing that curated bioactivity data scales test-loss 2.3× more efficiently than parameter count.
Scaling Laws Benchmarking
09
Kim, Sanchez, Zavoronkov et al. · Nat. Biotech. · May 10, 2026
Prospective case study across four targets integrating DEL screening, generative design, and automated SAR; median cycle time drops from 24 to 8 weeks.
Active Learning Automation
10
Segler, Coley, Jensen · Chem. Sci. · May 11, 2026
RL reward shaped by retrosynthetic tree cost; 78% of top-ranked compounds confirmed synthesizable by organic chemists vs. 34% baseline.
Retrosynthesis MPO

Social Media Pulse

Signal from researchers, executives, and investors.

Stock Monitor

AI drug-discovery pure-plays, platform adjacents, and the pharma buyers. Prices as of market close Friday. Click any ticker for live data.

Biggest Gainer
RLAY +6.4%
RLY-2608 Ph3 news
Biggest Loser
RXRX −8.1%
Guidance cut Q1
Highest Volume
03696.HK
3.2x avg volume
Sector Index
+0.8%
AIDD-20 composite

AI Drug Discovery Pure-Plays

Company Ticker Exchange Price Weekly Δ Market Cap Volume
Insilico Medicine03696.HKHKEXHK$42.60+4.2%$4.8B12.4M
RecursionRXRXNASDAQ$4.12−8.1%$1.1B18.2M
Relay TherapeuticsRLAYNASDAQ$5.87+6.4%$1.0B5.8M
SchrödingerSDGRNASDAQ$21.34−3.2%$1.6B2.1M
Tempus AITEMNASDAQ$58.90+2.1%$9.8B8.6M
AbCelleraABCLNASDAQ$3.21−1.8%$960M3.4M
AbsciABSINASDAQ$3.04−4.6%$380M1.2M
BenevolentAIBAI.LLSE£0.42+1.2%£90M0.3M
XtalPi / QuantumPharm02228.HKHKEXHK$6.88+3.7%$2.1B8.9M
EvotecEVT.DEXETRA€7.24−2.4%€1.3B4.1M
Neumora TherapeuticsNMRANASDAQ$8.44−2.8%$1.2B1.8M
Structure TherapeuticsGPCRNASDAQ$32.10+5.1%$1.8B1.4M

Pharma AI Buyers & Enablers

Company Ticker Exchange Price Weekly Δ Market Cap AI Relevance
Eli LillyLLYNYSE$842.10+1.8%$798BIsomorphic, Insilico, Iambic
Bristol Myers SquibbBMYNYSE$52.40−0.7%$106BNimbus ($6.1B), Schrödinger
NovartisNVSNYSE$114.20+0.9%$232BIsomorphic, Generate (acq.)
AstraZenecaAZNNASDAQ$76.88+1.5%$238BBenevolentAI, internal AI
RocheROG.SWSIXCHF 284.60+2.8%$240BRecursion ($150M), Genentech
PfizerPFENYSE$28.12−1.1%$160BInternal AI center
AmgenAMGNNASDAQ$304.88+0.6%$164BdeCODE Genetics ($14.3B)
RegeneronREGNNASDAQ$742.30+2.4%$80BGenetics Center, 500K exomes
Merck & Co.MRKNYSE$104.75−0.4%$265BAtomwise (acq.), internal
Gilead SciencesGILDNASDAQ$86.40+0.3%$108BInsitro partnership
SanofiSNYNASDAQ$52.18+1.1%$132BInsilico ($1.2B), Atomwise, BioMap
NVIDIANVDANASDAQ$924.50+3.2%$2.3TBioNeMo, Clara, GPU infra

Prices are indicative Friday close. Click ticker for live quote. Weekly Δ = Monday open to Friday close. Market caps rounded. Data refreshed every Sunday for Monday publication.

Thought Leadership

Analysis

Why 0-to-DC Data Is the New Moat in AI Drug Discovery

AI Pharma Weekly Research Team · May 11, 2026

The hardest dataset to assemble in drug discovery is not genomic, chemical, or clinical — it is the longitudinal record of what a molecule actually did from target nomination through to development candidate. Every failed assay, every SAR iteration, every off-target surprise, every pharmacology tweak — that is the tacit knowledge AI models need to compress timelines, and it exists almost nowhere as training data.

Insilico's 30+ internal DCs (development candidates) are a flywheel because each program generates tens of thousands of paired observations: compound → assay → decision → next compound. This is the opposite of public datasets like ChEMBL, which capture terminal bioactivity without the decisions that got there. Ranked against scaling-law evidence (see this week's Paper #08), it is increasingly clear that curated decision data beats parameter count for real-world property prediction.

Platform-only companies face a structural gap here. You cannot bootstrap target-to-DC data from licensed screens, academic literature, or simulated assays — by construction, those datasets lack the counterfactuals that matter. The companies that own both the generative stack and the wet-lab decision log will widen their lead each quarter. Everyone else is optimizing on the wrong distribution.

Read more →

Week Ahead

What to watch, May 11–17.

Mon
11
Week 19 recap published
Tue
12
AACR spring symposium
Wed
13
Recursion Q1 earnings
Schrödinger analyst day
Thu
14
FDA AI Guidance Workshop
Fri
15
Tempus AI investor update
Sat
16
JPM Healthcare side event (Boston)
Sun
17