Disambiguation of DNA-damage-response embedding clusters

Characterize the more specific phenotypes represented by the five distinct DNA-damage-response clusters in the Bayesian Probabilistic Matrix Factorization gene-embedding space.

Background

The paper analyzes the ten-dimensional Bayesian Probabilistic Matrix Factorization embeddings used to initialize AssayFormer and finds that their clusters align primarily with CRISPR-screen phenotype categories rather than canonical pathways or protein complexes.

Within this organization, five separate clusters are associated with DNA-damage response. The authors state that they could not further distinguish these clusters into more specific phenotypes, leaving their biological interpretation unresolved.

References

Notably, five distinct clusters are associated with the DNA-damage response, which we were unable to further disambiguate into more specific phenotypes.

— Biology-in-the-loop: Amortized Adaptive Hit Discovery in CRISPR Screens  (2609.11877 - Edwards et al., 10 Sep 2026) in Appendix, Section “Structure of the BPMF embedding space,” paragraph “HDBSCAN clustering”