We search targets, partners, payloads, sequences and settings against what your asset actually does, and hand back the few worth building. The model runs in seconds, so the search is wide enough to find the pairings nobody would think to try.
Biology runs on redundancy and feedback, so blocking one route very often means the cell reroutes through another. Which partner closes that route, in which setting, in which order, is where the outcome is actually decided. It is also the part no screen was built to answer.
The obvious indication is chosen for familiarity. Where a mechanism separates depends on the biology of the setting, and the setting closest to the last paper is rarely it.
Screens throw away the best partners. Sensitisers, immune primers and resistance-breakers carry weak signals on their own, so a pipeline tuned for single-agent efficacy discards them.
Resistance arrives on schedule. Durability erodes unless the escape routes are designed around from the start, which means knowing them before the disease finds them.
A full pairwise screen of 100 drugs at 100 doses runs to roughly 50 million experiments, and a dish still cannot see the immune compartment clearing a clone, the stroma walling a drug out, or what changes when you reverse the order of two agents. So we do it in a model that can, fast enough to search the whole space rather than a corner of it.
Built for Founders & CSOs · Discovery · Lifecycle Management · Franchise & TA leads · Portfolio Strategy
These are the factors that decide a real outcome and that a screen or a frequency-trained model leaves out. The distilled model carries all of them, which is what makes a search this wide worth running.
Discovery · Target ID
Rank pairs and triples of targets rather than single ones, scored on whether the combination closes the routes a tumour would otherwise take. In our first months we found that despite enormous genomic variability, resistance strategies recur across cancers in a similar order, which is what makes a search this size tractable.
Founders & CSOs · Portfolio Strategy
The setting where your mechanism separates first, the patient context that defines it, and the partner that widens the margin. Ranked by where the biology actually wins, with the experiments that de-risk the choice before you commit the programme.
ADC · LNP · viral vectors · cell therapy
A delivery platform can carry almost anything almost anywhere, which is the hard part. We search payload against target against indication using what your platform genuinely does well, and hand back a lead with a resistance map showing where a second cargo will be needed later.
Lifecycle Management · TA leads
The combination, sequence and setting that keep a marketed asset first-line while the backbone moves under it. Each candidate comes with a predicted survival benefit and the escape route it was designed to close.
Venetoclax blocks the survival protein BCL-2. Azacitidine pushes blasts into depending on it. In acute myeloid leukaemia that pair became a standard of care. In myelodysplastic syndrome, a blood cancer with largely similar biology, it deepened remissions and then missed on survival. Any argument from resemblance calls both the same way.
VIALE-A
Azacitidine + venetoclax · AML
Tumour populations over time
Overall survival
Priming converts. The blasts that azacitidine pushes into BCL-2 dependence are killed, the burden stays down, and the curves separate. Median OS 14.7 months against 9.6.
VERONA
Azacitidine + venetoclax · higher-risk MDS
Tumour populations over time
Overall survival
Priming happens and does not convert. Remissions deepen, the resistant fraction keeps growing underneath, and the curves close again. The overall-survival co-primary was not met.
Is the marrow primed and dependent on BCL-2, so that killing the primed fraction removes what drives the disease? In AML it is. In MDS the priming happens, and the cells that carry the disease forward are not the ones it kills.
Population plots are model output. Survival plots are drawn to the published read-outs: VIALE-A reported a median OS of 14.7 months against 9.6 (NEJM 2020;383:617); VERONA (NCT04401748) deepened remissions with the overall-survival co-primary not met, and is drawn to that shape rather than to published medians.
Send us a mechanism, an asset you want to defend, or a platform looking for its first indication. Because the model runs in seconds we can usually turn an initial review around within hours, and often for free, to show what can be done and to suggest a pathway to the optimal approach.
More than once a partner has come back somewhere between impressed and unsettled, because what we sent matched what they had spent a year working out in the lab, and included things they had not found yet. Those are the conversations we like having.