Programs by stage
Stage definitions, stated plainly. Target-to-hit means candidates have been designed and computationally scored against a target. In design means the program is being specified. Planned means work has not started. Candidates are validation-ready, with academic medical center collaborations being finalized. Open the Programs tab for indications, modality, and detail on each.
Every candidate in the pipeline was designed de novo by our engine and is a LillixBio asset. Candidates are validation-ready, with academic medical center collaborations being finalized.
Oncology targets engaged
Selected for genomic frequency across tumor types and for how completely the current standard of care fails the patients who carry them. Select any target for indications.
Four modalities, one design loop
The engine is not a small-molecule company with side projects. Modality is an output of the target, not a constraint of the company.
Every program, in full
Select any program to see indications, modality, and where it sits in development.
Our furthest-advanced program. MEK1 sits at the junction of MAPK signaling, where it drives proliferation across a large share of solid tumors. Candidates were generated de novo against the allosteric pocket and scored jointly for potency, selectivity, and synthetic tractability, with manufacturability treated as a design constraint from the first molecule rather than a downstream filter. Candidates for this program are validation-ready, with academic medical center collaborations being finalized.
KRAS G12D is the speed-to-proof program. It is the most common KRAS variant in pancreatic cancer and one of the hardest targets in oncology — which makes it the cleanest external test of whether the engine produces real binders against hard surfaces. Candidates for this program are validation-ready, with academic medical center collaborations being finalized.
RAC1 governs cytoskeletal remodeling and cell motility — the machinery tumors use to invade and metastasize. Targeting it addresses progression rather than bulk, and it is a recurrent driver in melanoma alongside activating BRAF and NRAS lesions.
β-catenin is the terminal node of Wnt signaling and a textbook undruggable target — a large, flat protein–protein interface with no classical pocket. It is exactly the class of surface generative design exists to attack, and the program spans both small molecules and engineered binders.
Mutant IDH1 produces an oncometabolite that rewires the epigenome. The biology is well characterized and the patient populations are genomically defined, which makes it a strong fit for a design engine that starts from a patient's own sequence data.
PIK3CA is among the most frequently mutated oncogenes in human cancer. The clinical problem is not whether it can be inhibited but whether it can be inhibited selectively enough to be tolerated — so isoform and mutant selectivity are scored as primary objectives, not afterthoughts.
AKT1 runs immediately downstream of PI3K and is the node through which much of that pathway's survival signal passes. Running both programs in parallel gives us vertical coverage of a single axis and a designed answer to the resistance that follows single-node inhibition.
Class II and III BRAF alterations are left behind by the approved V600E inhibitors — in some cases those drugs make the biology worse. These patients are genomically identifiable today and have no matched therapy, which is precisely the gap a design engine closes faster than a screening campaign.
Exon 20 insertions reshape the EGFR binding pocket so that standard TKIs no longer fit. The structural problem is well defined and narrow — a design brief rather than a discovery question — and the patient population is already routinely sequenced.
Further oncology targets are in active design beyond those listed individually. They span receptor tyrosine kinases, cell-cycle regulators, chromatin remodelers, and transcription factors — chosen for genomic frequency and for how badly the existing standard of care fails the patients who carry them.
Proprietary immune activation technology: rather than inhibiting a single node, we program the patient's own immune system to identify and destroy tumor cells. The platform is modality-independent of our small-molecule work and applies across solid tumors where checkpoint therapy alone has not delivered.
Resistant infection is the clearest demonstration that the platform is not an oncology company wearing a platform costume. The same design loop — read the biology, design against it, manufacture the result — runs against bacterial targets with no change to the underlying system.
For monogenic disease the defect is known, singular, and inherited — so the therapeutic goal is correction rather than suppression. Programs here are in design, with the engine scoring edit strategy, delivery, and off-target risk as one joint problem instead of three sequential ones.