XtalPi launches Kodexia, its platform for six siRNA programs

Why it matters

Kodexia extends XtalPi's physics-and-automation approach into siRNA. Its six internal programs give the founders a direct test of whether the platform's design-speed claims can yield drug candidates.

XtalPi, co-founded by Jian Ma, Shuhao Wen and Lipeng Lai, has put six proprietary siRNA drug programs behind Kodexia, a platform combining generative AI, biological modeling and automated lab work. The announcement gives XtalPi a new modality to build on its founders' original bet: use computational physics and experiments together to make molecular discovery more predictable. XtalPi calls Kodexia the industry's first platform of its kind. The test is whether its internal programs can produce candidates that advance beyond preclinical work.

The three co-founders came to XtalPi from physics research. XtalPi says the founders met as MIT postdoctoral researchers in 2015 and started with a specific problem: predicting crystal structures. Lai, who leads AI development, had also worked as a software developer at Epic Systems and completed postdoctoral research through the SUTD-MIT Graduate Fellows Program, according to XtalPi's 2024 annual report. Kodexia's core model, siFormer, applies RNA thermodynamics and structural features to sequence design, then sends the designs into experiments intended to feed results back into the models.

A platform built around its own pipeline

Small interfering RNA, or siRNA, aims to silence disease-related genes before they produce proteins. XtalPi says Kodexia combines sequence design and chemical modification with experimental validation, delivery design beyond the liver, and patent strategy. Its stated delivery research covers the kidney, spleen and adipose tissue, with antibody, peptide and small-molecule conjugates among the approaches under consideration. The platform also designs dual-target siRNAs, which are intended to act on two targets in one molecule.

The first announced programs are internal assets, spanning metabolic, renal, respiratory and central nervous system diseases. XtalPi singled out an IgA nephropathy program, saying it reached non-human-primate efficacy data seven months after initiation and showed greater activity and durability than a clinical-stage reference molecule aimed at the same target. XtalPi said the program was on track for preclinical candidate selection within nine months.

The platform's value will depend on its output, as well as its ability to generate sequences. XtalPi's August 19th interim results had already described Kodexia as an operating platform with tens of thousands of wet-lab data points and six siRNA programs; the same report said its most advanced program had reached the preclinical-candidate stage. The October 6th announcement formalizes the platform's public profile, while the underlying research and pipeline were already in motion.

XtalPi's interim report says it aims to earn through research services and infrastructure while advancing proprietary drug assets for potential co-development or licensing. In its August 19th interim results, XtalPi reported external AI4S deployments, including a compound-management system delivered to Eli Lilly, an autonomous drug-synthesis system delivered to JW Pharmaceutical, and an intelligent synthesis workstation deployed across 13 customers. The report does not identify a specific external Kodexia siRNA deployment. The portfolio lets XtalPi test the system across targets and indications, while any successful candidate could become an asset in its own right. XtalPi's filings describe that combination of service work and proprietary assets as part of its broader drug-discovery model.

Early benchmarks

XtalPi says its laboratories run more than 500 in vitro and 30 in vivo experiments each week. The October release reports a nearly threefold gain in molecular-design efficiency against conventional workflows and says more than half of first-round designs across multiple programs had stronger in vivo activity than positive controls. These are company-reported internal benchmarks; they do not establish that a drug will work in people.

The August interim report used different language for efficiency, saying Kodexia more than doubled research-and-development efficiency against conventional methods and improved molecular-property prediction accuracy by about 266%. It also reported that more than half of first-round molecules across multiple pipelines outperformed positive controls in vivo. The releases do not present the efficiency figures in a shared framework, so they should not be treated as directly comparable measures. Across both releases, XtalPi says its early designs have performed better than controls in its own preclinical tests.

XtalPi began by applying computational methods to crystal structures, then expanded into drug discovery and laboratory automation. siRNA requires a different mix of biology, chemistry and delivery work, especially for treatments aimed beyond the liver. XtalPi is applying its computational-and-experimental loop to that design problem, using its own pipeline to generate data and demonstrate speed.

The programs will show whether the IgA nephropathy asset advances on the timeline XtalPi has set and whether the other programs produce candidates worth developing or licensing. Their progress will determine the value of XtalPi's platform claims.