Generative Biology
Generate:Biomedicines
by Generate:Biomedicines, Inc.
Generative AI platform designing de novo protein therapeutics entirely from computational first principles
Category
Generative Biology
Founded
2018
Headquarters
Somerville, MA, USA
Overview
Generate:Biomedicines has built Chroma, a generative AI model for protein design that works analogously to image generation models like Stable Diffusion — but for three-dimensional protein structures. The Chroma model can generate novel protein structures and sequences conditioned on arbitrary properties such as binding specificity, thermostability, size, symmetry, and functional constraints, enabling the de novo design of protein therapeutics without starting from a natural protein scaffold. Pharmaceutical partners and Generate's internal pipeline team use Chroma to design novel protein therapeutics across antibodies, bispecifics, cytokines, and entirely new protein modalities. The company raised $273 million in Series B funding in 2022, one of the largest rounds in generative biology, to build a pipeline of de novo designed protein drugs that are physically impossible to arrive at through traditional antibody discovery or rational design approaches. Generate:Biomedicines differentiates through the generality of the Chroma architecture — unlike antibody-specific generative models, Chroma operates across the full space of designable proteins. The model accepts multimodal conditioning inputs including partial structures, binding partner coordinates, and sequence constraints, making it highly programmable for different therapeutic modalities. The company's ambition to design drugs that have never existed in nature distinguishes it from optimization-focused AI drug discovery platforms.
Key Features
Multi-Objective Optimization
Balance efficacy, selectivity, toxicity, ADMET properties, and synthesizability simultaneously.
Novel Molecule Generation
Generative models design molecules with desired properties including efficacy, selectivity, and synthesizability.
Synthesizability Assessment
Score generated molecules for synthetic accessibility and suggest practical synthesis routes.
Property Prediction Integration
Integrated property prediction validates generated candidates against multiple biological criteria.
Rapid Candidate Enumeration
Generate thousands of diverse candidates for experimental validation in hours.
Pros & Cons
Pros
- +Cross-domain generative capabilities span small molecules, peptides, proteins, and nucleic acids
- +Generative models design novel molecules, proteins, and genetic sequences with desired properties
- +Multi-objective optimization balances efficacy, selectivity, toxicity, and synthesizability simultaneously
- +Transfer learning from large biological datasets enables design in low-data domains
- +Inverse design capabilities specify desired functions and generate candidate sequences automatically
- +Rapid iteration cycles generate thousands of candidates for experimental validation in hours
- +Interpretable models reveal structure-function relationships driving design decisions
Cons
- −Interpretability of generative model decisions remains limited for regulatory submissions
- −Computational costs for training and inference of large generative models can be substantial
- −Synthesizability of generated molecules is not always guaranteed by the model
- −Generated designs require experimental validation — computational predictions don't guarantee function
Use Cases
Research Workflow Optimization
AI-powered optimization of research workflows to accelerate discovery timelines and improve reproducibility.
Data Analysis & Insights
Machine learning analysis of complex biological datasets to extract actionable insights and identify patterns.
Collaboration & Knowledge Management
Platform-enabled collaboration across distributed research teams with integrated data sharing and knowledge capture.