Beyond AI Drug Discovery: Astromech and the Rise of Predictive Biology

For the past several years, the dominant story in AI and life sciences has been drug discovery. Companies have raised billions of dollars to apply machine learning to protein folding, molecular design, and clinical trial optimization. The pitch is faster, cheaper drug development using AI to do in months what once took decades.

Astromech is not building a drug discovery company. The Dallas-based startup, which just closed a $20 million funding round at a $3.8 billion valuation, is pursuing something earlier in the research chain and considerably broader: a system that can predict how living systems change before those changes occur.

The distinction helps clarify Astromech’s approach. While drug discovery is one potential downstream application, predictive biology could provide an earlier foundation for such work.

What Is Predictive Biology?

Much of computational biology focuses on present-day data, with models analyzing current genetic sequences, gene expression, and observable traits. The goal is to understand what a biological system is doing right now.

Astromech's approach is different. The company studies how biological traits developed across evolutionary time, using that longitudinal record to project where systems are likely to go next. Co-founders Ben Lamm and George Church describe it using the analogy of weather forecasting: a forecast works because meteorologists can study the historical behavior of atmospheric systems, not just current conditions. Astromech aims to bring the same forecasting logic to biology. "Biology runs the world, and historically, we have only reacted to it," said Lamm. "We can describe biology in extraordinary detail, yet we still struggle to anticipate what comes next."

How Astromech Models Biological Change

Astromech’s platform draws on three layers of input: genomic data from living and extinct species, evolutionary data capturing deep-time ancestry and divergence, and functional data including gene expression, trait variation, and responses to environmental change. Astromech combines these to train on what it describes as 3.8 billion years of biological history.

Two computational engines power the architecture. A deep learning suite finds patterns across species, learning how genes are expressed in different organisms, how vulnerabilities emerge over generations, and how different lineages responded to similar pressures. A second engine works backward through evolutionary history using ancestral state reconstruction, then applies the same mathematics forward to project where a system goes next. The two engines are fused into a unified modeling framework.

The technical core extends ancestral state reconstruction beyond DNA sequence. Many phylogenetic methods infer ancestral proteins at evolutionary tree nodes. Astromech reconstructs ancestral regulatory state: chromatin accessibility, gene expression, and functional annotation across data types, integrated through a Bayesian framework that produces calibrated probability estimates rather than single-point predictions. The platform does not just report a predicted outcome; it assigns confidence levels to it.

For many of the most complex and consequential biological traits, the important variation occurs outside protein-coding regions. Longevity, cancer resistance, immune function, and stress tolerance are all heavily regulated traits where sequence differences between species often lie in how genes are switched on and off, rather than in the proteins themselves.

"Most of the variations that matter for complex traits are regulatory rather than coding," said Church. "Reconstructing the ancestral regulatory state, not just the ancestral protein, has the crucial explanatory power. That takes functional data across many species rather than sequence alone, and AI reconstruction is cheap enough to run genome-wide. Neither was true ten years ago."

This is why training on a single reference genome is insufficient for Astromech's goals. Models built on one or two reference species can help researchers study variation within a lineage but may offer a more limited view of regulatory changes across lineages. Astromech trains across species and supplements public datasets with functional data generated in-house, building a comparative foundation that allows the models to study what changed, when it changed, and what effect those changes had.

Astromech's platform is designed to generate three types of output: where a genome or population is likely headed, where a biological system is most vulnerable to breakdown, and which regulatory mechanisms are driving those trajectories.

Potential Applications Across Multiple Fields

The range of potential applications reflects how broadly biological prediction applies. In human health, the platform could map disease risk and the genomic and regulatory drivers of aging. In biosecurity, it could flag susceptibility to pathogens across species before they reach human populations or identify drug resistance trajectories before they become treatment failures. In agriculture, it could model herd vulnerability under disease and climate stress. In conservation, it could identify which ecosystems are most at risk under changing environmental conditions.

Longevity as an Initial Test Case

Astromech's initial public demonstration maps 46 longevity-associated genes across a time-calibrated tree of life. Choosing longevity as a first application reflects both scientific opportunity and the platform's comparative advantages.

Long-lived species present natural experiments in biological resilience. Bowhead whales can exceed two centuries of life despite body mass that would ordinarily predict elevated cancer risk. Brandt's bats weigh only a few grams but can live over 40 years, far outpacing mammals of comparable size. Asian elephants have evolved cancer-suppression mechanisms strong enough to protect them despite exceptional longevity. Birds routinely outlive mammals of similar size despite higher metabolic rates.

Each of these represents an independent evolutionary solution to a shared biological challenge. By reconstructing when those solutions emerged, on which lineages, and what genomic and regulatory changes drove them, Astromech aims to generate more precise hypotheses about the mechanisms underlying resilience and aging. Those hypotheses can then inform downstream research programs.

According to Astromech, retrospective validation of its pipeline recovered trait-associated genes established in published research and identified additional candidates. Prospective validation through partner pilots is the next phase.

Building an Upstream Research Tool

Predictive biology, as Astromech is building it, is not a replacement for drug discovery, clinical research, or experimental biology. According to Astromech, the platform is designed to generate better-informed hypotheses earlier in the research process, flag potential biological risks, and surface mechanisms for further experimental validation.

The company's ambition is to build infrastructure that sits upstream of the industries that depend on understanding biological change, adapting the same underlying framework to different traits, species, and applications as validation progresses. Whether predictive biology becomes a recognized category will depend on whether the forecasts hold up. Astromech is betting that they will.

This article is for informational purposes only and does not substitute for professional medical advice. If you are seeking medical advice, diagnosis, or treatment, please consult a medical professional or healthcare provider.

BDG Media newsroom and editorial staff were not involved in the creation of this content.