Can the binding affinity for new drugs be algorithmically determined?
About the Client
Background
DNA-seq is a computational drug discovery company focused on identifying non-conserved, mutation-prone residues in kinase structures and linking those signatures to drug resistance.
Overview
Invene built a Claude-powered harness that lets computational chemists query the Protein Data Bank in plain English and orchestrate analysis scripts as agent tools — eliminating API expertise as a prerequisite for day-to-day drug discovery research. Their scientists — including computational chemist Davide Moiani and structural biologist Janusz Sowadski — needed to work fluidly with protein structure data from public databases and run a growing library of custom analysis scripts.
Challenge
Computational chemists used the Protein Data Bank (PDB) for drug discovery, but its 3,000+ pages of API documentation made complex queries difficult without specialized expertise. Critical analysis functions were also scattered across dozens of standalone scripts, forcing researchers to manually select the right tools. Early AI tool-calling approaches were inefficient and unreliable, increasing token usage and producing inconsistent results. The team needed a solution that simplified PDB access, intelligently coordinated analysis workflows, and integrated results into an interactive 3D molecular visualization environment.
80%
5
100%
Solution
Invene led both the AI engineering architecture and full-stack delivery of the solution. The team developed a retrieval-augmented generation (RAG) layer that translates natural language requests into accurate PDB API queries, a context-aware orchestration engine that dynamically manages tool availability based on the researcher’s workflow, and a script harness that enables existing Python analysis tools to be exposed as on-demand agent capabilities. To support long-term scalability, Invene also implemented an Entity-Component-System (ECS) architecture within the IDE, allowing new scripts and data sources to be added with minimal effort. Using a Shape Up delivery approach with staggered releases, the team delivered a working prototype within the first development cycle and completed the full platform build by December 2025.
Results
The solution transformed how DNA-seq researchers access and analyze protein structure data, enabling natural language queries across PDB and UniProt without requiring API expertise for routine discovery tasks. Existing computational chemistry scripts were unified within a single AI-powered IDE, where tools are intelligently invoked based on context rather than manually selected. Delivered on schedule with a prototype released in August–September 2025 and the full platform completed in December 2025, the system remains in active daily use more than six months after launch. Continued product evolution and discussions around a follow-on engagement demonstrate strong user adoption, while the dependency graph and context-management approach developed by Invene has become a reusable framework for future AI agent implementations.
More Case Studies
A 3D printed component that turns a scuba mask into a reusable N95.
Can healthcare providers accelerate patient access platform migrations at scale?
Can a member-facing web portal be built in 8 weeks?
Transform Ideas Into Impact
Discover how we bring healthcare innovations to life.
Tackling the Hardest Software & Product Problems in Healthcare
Privacy Policy
