Introducing Infera and BioBase

At Infera, we're building AI agents to operate scientific instruments and analyze the data they produce. Over the past several months, we've been developing instrument-control software, building evaluations for scientific tasks, and expanding the tools available to our agents.

Today, we're announcing BioBase in open beta: a desktop research environment where scientists direct AI agents to investigate scientific questions and organize data from laboratory equipment. BioBase enables scientists to unify their hardware, research packages, and agent control under one roof.

Research with BioBase

One of the greatest challenges in scientific research today is the complex web of interactions between hardware interfaces, software tools, and the scientist. Indeed, scientists routinely grapple with transporting data across devices, schemas, and folders to inspect their work and investigate their hypotheses.

In BioBase, researchers describe the work they’d like performed; then, BioBase spins up several agents to plan the approach and write scripts to carry out the task requested. These agents query hundreds of scientific databases and delegate autonomously to subagents, asking the researcher for decisions as needed. The resulting files return to the project, neatly sorted in the active directory.

Moreover, BioBase records the steps and commands from an analysis alongside its code and output files, so the researcher can inspect the traces that support a given outcome. Researchers can stop a task and revise its instructions before continuing in the same project.

In visualizing scientific research, BioBase presents the findings alongside the underlying measurements and references, with interactive figures and molecular structures that can be examined alongside the source data as researchers develop their work.

Natively shipping high-performance computation

Some steps in a study need a graphics processing unit (GPU), such as running a protein language model over a set of sequences or training a model on screening data. BioBase agents send those steps to GPU sandboxes remotely, choosing a GPU compute profile for the step that needs one while the rest of the analysis continues on CPU. For structure prediction, docking and protein and small-molecule design, agents also call NVIDIA BioNeMo models, including Boltz-2, DiffDock, RFdiffusion and ProteinMPNN.

For many scientists, accessing an HPC remotely requires scheduling, sharing, and negotiating with other teams for cluster time. On Infera BioBase, agents are equipped with the ability to undertake GPU-intensive or high-performance computing on the same budget as inference.

GPU steps that once meant requesting an HPC allocation, logging in to a cluster and copying data onto it now run from the BioBase desktop app, and agents working in parallel each run their own sandboxes at the same time. For the first time, labs can invest grant funding into HPC work on-demand, without having to invest in volatile, depreciating graphics processing assets.

Driving results directly from hardware

Today, raw instrument data still takes manual work to turn into scientific answers. Measurements need to be connected to sample information and processed with the lab's methods before researchers can interpret them. That computational work continues as researchers refine an analysis or ask a different question of the same data.

As part of Infera Bio’s mission to develop technologies that close the loop between in silico work and the wet lab, BioBase is the first research environment that allows scientists and enterprises to seamlessly integrate downlink and experimental data from their instruments into their R&D pipelines.

Our ambition is for scientists to direct entire drug discovery programs in BioBase, from target assessment through experimental follow-up. We're developing agents to operate instruments and robots alongside BioBase, with the aim of connecting experimental execution to the pipelines that analyze the resulting measurements. The goal is to create a fully integrated research environment where agents carry out experimental procedures as well as computational studies, returning results that scientists can use to cure disease faster.

Drug discovery on BioBase

For applications in drug discovery, BioBase connects target assessment with experimental compound analysis, supporting discovery research from the first principles behind a given target through the interpretation of laboratory screens.

Over the past few months, scientists have used BioBase to retrieve target-disease associations and binding measurements, while examining evidence for pursuing a target alongside compounds already studied against that target, ultimately having much greater confidence in defining the scope of experimental programs.

In concentration-response experiments, scientists developed pipelines to fit curves to laboratory measurements and return potency estimates, while also autonomously examining plotted measurements and fitted responses together and comparing assayed compounds to generate follow-up experiments.

By combining target research with the analysis of experimental pharmacology, BioBase gives discovery teams a computational environment that extends from the initial assessment of a target into successive rounds of compound testing.

Availability

BioBase is available in open beta for macOS, Windows, and Linux, with free and paid plans. Downloads are at infera.bio/biobase.