Variant Bio announced the launch of Inference, the world's first agentic genomic drug discovery platform. Built on seven years of development and over USD 100 million of investment in data and infrastructure, Inference transforms how pharmaceutical R&D teams harness human genetic data to discover and validate therapeutic targets.

The company developed the platform to be the definitive solution for genomic drug discovery. As a company that generates proprietary human genomic data from studies around the world, processes and analyses massive multi-omic datasets and advances its own drug development programmes, Variant Bio is positioned to build and continuously improve this platform through direct application to real-world drug discovery challenges.

The company has ingested and harmonised data from major genetic and multi-omic association studies and numerous global biobanks, alongside proprietary datasets from its studies with under-represented populations. The platform has processed hundreds of billions of genetic associations across tens of thousands of genome-wide association studies, creating the largest single unified source of public and proprietary human genomic data for drug discovery.

The platform's architecture reflects a fundamentally different design philosophy. Rather than offering static, pre-computed results, Inference provides structured and harmonised data instantly available for dynamic downstream analyses. A proprietary Application Programming Interface (API) powers the entire platform, enabling instant retrieval of gene and variant-level data across hundreds of millions of data points in seconds.

"With Inference, we built the tool we wished existed when we started Variant Bio. Enterprise software design principles, applied by people who actually do genomic research. When you're solving your own problems, you build something real,” said Stephane Castel, Co-Founder and Chief Technology Officer, Variant Bio.

Inference incorporates two breakthrough AI capabilities designed to accelerate target discovery. VB-Predict creates molecular models of disease using proprietary transcriptomic, proteomic and metabolomic data and neural networks. These models can be applied across cohorts to predict traits and identify the targets most correlated with disease outcomes.

VB-Atlas learns embeddings from tens of thousands of association studies to map genome-wide relationships between genes and phenotypes. This enables instant querying of genes associated with any phenotype or genes with similar profiles, and critically, makes massive datasets immediately accessible to Large Language Models (LLMs).

Most significantly, Variant Bio is introducing Inference Agents – autonomous AI agents capable of executing advanced analyses and interpreting results. These agents use LLMs to plan and carry out analyses requiring hundreds of chained data queries and API calls – from validating a therapeutic target by assessing causality, determining effect direction and scanning for safety liabilities, to systematically ranking putative causal genes across hundreds of disease-associated loci by integrating across evidence types and datasets.

The framework is built on emerging standards including the Model Context Protocol (MCP), enabling seamless integration wherever teams interact with generative AI, including Claude and ChatGPT desktop applications. Unlike literature-review approaches, Inference Agents make novel discoveries from foundational data using first principles.

"As early AI investors, we've seen countless approaches to drug discovery. What Variant Bio has built stands out as one of the most impressive applications of agentic AI. They're avoiding artificial hype and building autonomous systems doing actual scientific research, with real implications for patients,” said Josh Wolfe, Co-Founder and Managing Partner, Lux Capital.

While Variant Bio initially built the platform for internal use, discovering targets that are the basis of four drug development programmes, the company is now opening Inference to research partners. The platform serves as the centrepiece of collaborations with multiple pharmaceutical partners, including multi-year discovery alliances with combined deal values exceeding USD 200 million.

Partners can upload their own proprietary genetic and multi-omic data to the platform, with a robust permission system ensuring data remains private and secure. Variant Bio continues to expand the data available on the platform, incorporating additional public and proprietary datasets on an ongoing basis.