Drug resistance prediction
Anticipate drug resistance before it emerges.
Topologix is developing a platform that predicts which mutations in a drug target will cause resistance to a compound, from the protein sequence and the compound’s structure. We are building it for drug-discovery teams and CROs working in oncology, infectious disease and antivirals.
Resistance is usually found too late
Mutations in the drug target are one of the main ways tumours, viruses and bacteria escape a therapy. A single amino-acid change can cut a compound’s potency tenfold or more while leaving the protein fully functional.
Most programmes learn which mutations matter from resistant patient samples, from clinical isolates, or from mutant panels run late in optimisation. By then the chemistry is hard to change. Measuring every candidate compound against every plausible mutant is slow and expensive, so most combinations are never tested at all.
What we are building
- Mutation-level resistance scoresFor each mutation and compound, the probability that the compound loses at least tenfold potency against the mutant compared with the wild-type protein.
- Compound and series triageRanking candidate compounds and chemical series by how well they are predicted to hold up against known and plausible mutations.
- Assay prioritisationPointing teams to the mutations and compounds most worth measuring first, so mutant panels are spent where they are most informative.
The predictions draw on three kinds of information: how the amino-acid change alters size, charge and chemistry at that position; how tolerated the change is across evolution, estimated with a protein language model; and the structure of the compound. More on the approach and how we evaluate it.
Early results
Our first model predicts resistance for compounds it has never seen. On a ChEMBL-derived corpus of 10,983 mutant and wild-type activity pairs across 215 target proteins, with whole chemical series held out of training, it reaches AUROC 0.827 ± 0.021.
The evaluation was pre-registered: comparisons and pass marks were fixed before any model was trained. A control that knows only which protein a pair belongs to scores 0.50, and performance falls to chance when the labels are shuffled. These results cover new compounds against targets in the training data, most of them kinases and HIV-1 reverse transcriptase.
Who it is for
- Medicinal chemistry teamsChoosing between series against a target where resistance is known or expected.
- Computational chemistry and CADD groupsAdding a resistance estimate next to potency and ADMET predictions when triaging compounds.
- Translational and clinical teamsDeciding which mutant assays to run, and anticipating which mutations may appear in patients.
- Contract research organisationsOffering resistance profiling alongside existing computational and assay services.
Therapeutic areas we are building towards: oncology, infectious disease and antimicrobials, and antivirals.
Become a design partner
We are working with a small number of drug-discovery teams and CROs to decide what Topologix should do first. If resistance has cost you a programme, or you are trying to anticipate it in one, we would like to hear how you handle it today and what a useful prediction would look like.
Design partners can also run a blind test on their own data: send structures and mutations, keep the measurements, and score our predictions yourself. We are happy to sign an NDA first.