AI-First Biotechs Turn to Illumina for Data: Newsletter, July 21, 2026

Freenome goes public, Tempus acquires Personalis, Broad/JAX team up on genome editing, Doudna lab embraces AI, and more.

A trio of AI-first drug discovery companies have signed on to Illumina’s Billion Cell Atlas (BCA) data offering, a sign that its bet on single-cell perturbation data is beginning to pay off.

On July 16, Illumina announced that Formation Bio, a startup that has raised more than $600 million, had joined the likes of AstraZeneca, Merck, and Eli Lilly in gaining access to data of millions of cells that had been run through perturbation screening experiments.

Illumina declined to disclose the other two, however, a spokesperson said “they are also in the AI drug discovery biotech space.” Financial and other details of the deals were also not disclosed.

I wrote about Illumina’s dive into generating single-cell perturb-seq data that might be used to train AI in late 2025 for GenomeWeb. At the time, the effort to capture how genetic perturbations affected cells across 200 different cell lines seemed to me like an add-on offering to pharma giants already engaged with Illumina’s whole-genome sequencing data through the Alliance for Genomic Discovery.

By landing companies like Formation Bio, which have been founded around the potential for AI models to accelerate drug development, Illumina is demonstrating that it has appeal to a different demographic.

In an email, a Formation Bio spokesperson said that Illumina’s BCA was the “best available harmonized, consistent dataset.” Larger perturbation screening data are “intractably expensive at scale” while publicly available data contain limited cell-type diversity and batch variability.

While Illumina also plans to offer access to AI models, the data are the “primary draw” and complement Formation’s data from electronic health records, the company said. In general, Formation’s approach is to acquire drug candidates already near or in clinical trials and use AI to make sharper decisions about which drugs to advance and in which patient populations.

Formation’s interest also suggests that genomic data could play more of a role outside of target discovery. The clinical assets it targets are “typically two to four years from a major value inflection (such as a Phase 2 readout) and often ones that have stalled, lacked resources, or been deprioritized elsewhere,” Formation said.

It’s not the only company that has suggested using AI and genomic data to address this phase of the drug development pipeline, but finding better uses for drugs that have already passed safety studies could present tremendous value to pharma companies.

Illumina noted that it has already sequenced 350 million cells. Whether companies will also be interested in the models Illumina plans to train on these data remains to be seen.

---

Other genomics news

Liquid Biopsy Company Freenome Goes Public

Freenome has completed the special purpose acquisition company (SPAC) deal announced in December and is now publicly trading on the Nasdaq under ticker symbol “FRNM,” reports Kelsy Ketchum of GenomeWeb. The company offers sequencing-based analysis of methylation on cell-free DNA for early cancer detection.

Parse Biosciences, a single-cell sequencing technology company now part of Qiagen, announced that the Garvan Institute of Medical Research’s genomics core facility has joined its certified service provider program.

“We see a growing number of requests for Parse projects and find the technology easy to implement and run, generating great data,” Chris O’Keeffe, cellular genomics lead at the Garvan Institute, said in a statement.

My Genomics Reading List

Papers I flagged this week across spatial transcriptomics and cancer biology:

Researchers from Xaiowei Zhuang’s Harvard lab introduced a new in situ RNA amplification method combined with MERFISH to achieve whole-transcriptome, isoform-resolved spatial transcriptomics in tissue, imaging roughly 33,000 distinct RNAs, including around 10,000 isoforms, in the mouse brain.

Cohen et al. (Xiaowei Zhuang’s lab). “Whole-transcriptome-scale isoform-resolved spatial imaging of single cells in tissues.” Cell. July 16, 2026.

Wu et al. “Single-cell and Spatial Omics Reveals Region-Specific Plasticity and Therapeutic Vulnerabilities in Metastatic High-Risk Neuroblastoma.” Science Advances. July 17, 2026.

Putzbach et al. “Single-cell transcriptomics shows how tumors reprogram neutrophils, exposing a targetable vulnerability that suppresses metastasis.” Science Advances. July 15, 2026.

Elsewhere on Substack

David Crean, a longtime biotech executive and venture capital investor, has written a sharp critique of how the life sciences funding ecosystem is working as designed but still failing patients.

The essay is mostly about the pharma side of biotech, but it highlights some of the real frustrations of trying to address health in a system that is primarily about making money.

Is Venture Capital Broken in Life Sciences?

Tidbits worth digesting:

“The first quarter of 2025 marked the lowest quarterly level of new biotech startup formation in the United States in at least a decade, the fourth consecutive year of decline.

Rare disease funding is a rounding error against need. Just 2% of all biotech seed and Series A dollars went to rare-disease companies in 2023, against a population of roughly 30 million Americans, half of them children.

Women’s health sits in the same place. Only about 2% of healthcare VC funding has historically gone to women’s health, against a market projected to reach $103 billion by 2030, and in 2020 only 5% of global R&D funding went to women’s health research. Founders report that endometriosis programs attracted investor interest only when reframed around fertility, which signals an underwriting bias rather than a data gap.

Elsewhere on the Internet

If you see anything you think would make a good newsletter item, let me know!