Predicting and blocking how cells respond: a repair blocker, a zero-shot drug model, and a lineage toolkit

A repair pathway, interrupted

In a preprint, researchers report that homoharringtonine, abbreviated HHT, binds directly to the XPC protein, a factor involved in DNA damage repair. The reported binding affinity is a Kd of 158.4 nM, and the binding depends on the Thr795 residue of XPC. m:…7113

The dependence on Thr795 narrows the reported interaction to a specific site on the protein: the preprint reports HHT binding to XPC as dependent on that residue, rather than as a general association with the protein as a whole. m:…7113

That binding matters because it disrupts the interaction between XPC and damaged DNA. The preprint reports that the disruption impairs two repair routes: nucleotide excision repair, known as NER, and non-homologous end joining, known as NHEJ. In other words, the compound does not damage DNA itself in this account; it interferes with the machinery a cell would use to repair damage. m:…7113

The practical consequence reported in the preprint is sensitization: HHT sensitizes tumor cells to platinum-based chemotherapy in both laboratory cell models and animal models. The finding therefore remains a laboratory and animal-model result, not a treatment claim. m:…7113

Predicting a response before measuring it

A second preprint tackles a different part of the same problem: what a drug will do in a cell type where its effect has not been measured. Its method, called GRAFT, enables zero-shot prediction of drug-induced transcriptional responses across cellular contexts by leveraging genetic perturbation data. The idea is to transfer information from one kind of perturbation, genetic, to another, chemical, so that a drug response can be predicted in a context where no direct drug measurement exists. m:…2217

The scale of the accompanying resource is large. GRAFT-Atlas comprises 102,128 predictions for drug-context combinations across 13 cellular contexts and 7,856 drug catalog entries. That breadth is the point: a researcher working in one cell type could consult predicted responses across many drug-context pairs rather than starting from no measurement at all in that context. m:…2217

The preprint also describes a refinement step for cases where a few measurements are available. GRAFT-Adapt refines predictions of unmeasured responses using a small number of drug measurements from the target context. The method therefore sits between two positions: prediction with no target measurement, and prediction adjusted by a limited amount of target data. m:…2217

Tracing where a cell came from

A third preprint addresses a gap in single-cell analysis: a cell's current measurements say little, on their own, about its history. Pycea, described as a toolkit in the scverse ecosystem, enables integrated analysis of lineage trees and cellular measurements, so that a cell's ancestry and its present state can be examined together rather than in separate workflows. m:…0862

The technical change that makes that integration possible is a data format extension. TreeData extends AnnData to store lineage trees alongside cellular measurements, eliminating the need for format conversion between lineage information and measurement data. m:…0862

Read together, the three preprints describe response as something with a past, a present mechanism, and a predictable future. Pycea traces the past, HHT blocks a present repair mechanism, and GRAFT predicts a future response in an unmeasured context. m:…0862 m:…7113 m:…2217

Keeping lineage trees and cellular measurements in one analysis, rather than converting between formats, is the practical point of the toolkit. Pycea's integrated analysis within the scverse ecosystem means a researcher can examine a cell's ancestry and its measurements together, and TreeData's extension of AnnData is what allows the tree to sit alongside the measurements in the first place. m:…0862