Molly Gibson, PhD, has focused her attention on building companies at the intersection of AI and life sciences for the past decade, having co-founded both Generate: Biomedicines and Lila Sciences.

Generate is an AI-based protein therapeutics company whose most advanced AI-designed antibody program entered Phase III clinical trials last year. Lila closed out a $350 million Series A last fall to build fully autonomous research labs driven by “scientific superintelligence.” Both companies are backed by Flagship Pioneering, where Gibson is an origination partner.

Gibson has now taken the reigns as CEO of her own enterprise to solve one of the most challenging problems in drug discovery: generating novel chemistry for targets that have traditionally been deemed undruggable.

Expedition Medicines unveiled from stealth in October with a $50 million commitment from Flagship. The new start-up will apply AI to significantly expand targets for small molecule drugs, where successful hits remain contingent on deep and highly structured protein pockets. This traditional approach misses many sensors, regulators, and transcription factors that drive disease through interactions spread across surfaces and lack grooves for molecules to grab hold.

“It felt like the right next step to apply my learnings from Generate and Lila to take small molecules to the same level that we’ve seen other technologies reach,” Gibson told me during a visit to Expedition’s Cambridge-based headquarters.

She says therapeutic modalities directly rooted in biology are “much further along” the path to programmability. As an example, companies like Generate demonstrate advances in protein therapeutics, where sequence, structure, and function can be deliberately designed rather than discovered.

“Small molecules have historically been more challenging for generative AI, but I think we are at an inflection point, with the right chemistry insights, data, algorithms, and compute finally coming together,” Gibson continued.

Expedition will develop drugs that bind shallow pockets by learning the rules of permanent bond formation, known as covalent chemistry. The approach contrasts from many of today’s AI models which use the 3D atomic positions of molecules to target deep pockets through reversible interactions. Gibson foresees AI to drive a revolution in quantum chemistry for small molecules, similar to how Nobel Prize-winning AlphaFold fueled a turning point for protein folding and design.

Expedition has spent three years in internal development and now seeks to demonstrate clinical proof points of its mature platform. The start-up has already solidified a partnership with Pfizer to identify target molecules correlated with prostate cancer disease progression and treatment resistance.

Fear no more 

Covalent chemistry has occupied a longstanding pillar in therapeutic development with many common drugs, including Tylenol, aspirin, and penicillin, all leveraging covalent mechanisms. Cancer-treating covalent kinase inhibitors, such as Ibrutinib and Afatinib, have seen extraordinary commercial success, with Ibrutinib reaching blockbuster status.

Yet, permanent bond formation has historically triggered fear, says Dean Stamos, PhD, co-founder and CSO, proteomics and chemistry at Expedition, in an interview with GEN Edge. Highly reactive candidates that modify the wrong substrates could lead to damaging immunogenic events.

Stamos asserts that machine learning powered by high throughput proteomics could bring a sea change that “designs in” the benefit and “designs out” the fear.

Aligned with this mission, Expedition applies AI models to generate safe and stable drugs that remain inert inside the body until activated by the right protein catalyst to form an irreversible bond.

The tech stack uses a mass spectrometry data engine that measures the potency of each small molecule against 20,000 sites in the proteome at amino acid resolution, offering a significant increase in scale compared to traditional labor-intensive screens.

These fit-for-purpose datasets provide an advance upon DNA-encoded libraries (DELs), where data is riddled with significant noise that diminishes the predictive power of AI models.

When pharma partners question whether covalent molecules can be both reactive and specific, Nate Stebbins, PhD, co-founder and chief strategy and operations officer at Expedition, answers with data showing that Expedition’s drug candidates can exclusively hit one target out of thousands of sites in the proteome.

This selectivity is achieved by a “distinct flavor” of covalency that uses fast chemical reactions driven by perfect molecular overlap of electrons, a contrast from traditional approaches that require drugs to wait in deep pockets for irreversible covalent events to occur.

“That brand of covalency is one of the only ways that you can reach into shallow pocket territory,” Stebbins told GEN Edge.

Expedition’s vision does not stop at undruggable targets. Looking ahead, the company’s proteomics platform opens opportunities to explore new modalities, such as proximity events that drive protein degradation or stability.

“Any technology that changes the rate at which we can create potent small molecule chemistry is going to dictate the future of medicine,” says Stebbins. “The thing we’re building is the best shot at unlocking generative chemistry.”