AI Drug Discovery
Relay Therapeutics
by Relay Therapeutics, Inc.
Harnessing protein motion to discover medicines against previously undruggable targets
Category
AI Drug Discovery
Founded
2016
Headquarters
Cambridge, MA, USA
Overview
Relay Therapeutics was founded on the insight that protein function is governed not just by static structure but by dynamic motion — the conformational changes proteins undergo as they interact with ligands, other proteins, and cellular machinery. The company's Dynamo platform integrates molecular dynamics simulation, cryo-EM, NMR, and machine learning to model how proteins move and to identify cryptic binding pockets that are invisible in static crystal structures. Drug discovery scientists at Relay use the Dynamo platform to design highly selective small molecule inhibitors against targets that have historically been considered undruggable. The company's lead program, RLY-2608 (a PI3Kα mutant-selective inhibitor for breast cancer), entered Phase II clinical trials and demonstrated tumor responses in patients who had progressed on prior PI3K therapy — validating the motion-informed design approach. Relay's differentiation lies in motion-based drug design, a category it effectively pioneered. The integration of experimental dynamics data (from specialized NMR and cryo-EM techniques) with computational simulation gives Relay a mechanistic depth that pure in silico approaches lack. The company has multiple oncology programs in the clinic and a growing pipeline targeting kinases, phosphatases, and transcription factors.
Key Features
Clinical Trial Prediction
AI models predict clinical trial success probability based on preclinical data and historical trial outcomes.
Multi-Target Optimization
Simultaneously optimize drug candidates across multiple biological targets for polypharmacology approaches.
ADMET Profiling
Comprehensive in silico prediction of absorption, distribution, metabolism, excretion, and toxicity profiles.
De Novo Drug Design
Design entirely new drug molecules from scratch using generative AI trained on billions of molecular interactions.
Binding Affinity Prediction
Deep learning models predict drug-target binding affinities with near-experimental accuracy.
Pros & Cons
Pros
- +Strategic pharma partnerships validate platform capabilities with billion-dollar deal values
- +AI-powered virtual screening accelerates hit identification by 10-100x compared to traditional high-throughput screening
- +Proprietary biological datasets spanning petabytes of experimental data enable novel target discovery
- +Closed-loop integration of wet-lab experiments with AI models continuously improves prediction accuracy
- +Multi-target drug discovery platform identifies candidates across oncology, rare diseases, and infectious disease
- +Foundation models trained on billions of molecular interactions predict drug-target binding with high accuracy
- +Reduces preclinical development timelines from years to months with computational candidate optimization
Cons
- −Computational predictions still require extensive wet-lab validation before clinical advancement
- −Black-box nature of deep learning models creates interpretability challenges for regulatory submissions
- −Long sales cycles and custom integration requirements extend time to value for new customers
- −Enterprise pricing accessible only to large pharma — prohibitive for academic labs and small biotechs
Use Cases
Virtual Screening & Hit Identification
AI-powered virtual screening of billion-scale compound libraries to identify drug candidates in days instead of months.
Target Identification & Validation
Machine learning models identify novel drug targets from multi-omics data and validate their therapeutic potential.
Lead Optimization
Computational optimization of lead compounds for potency, selectivity, ADMET properties, and synthesizability.