arXiv subjects
Publications and source records attributed to Pavlo Lutsik.
Revised following peer review. We expanded baseline comparisons, corrected evaluation leakage and read-boundary handling, clarified the confidence-weighted loss, and added sensitivity analyses for pooling and region selection. We also expanded TCS failure-mode and limitations analyses, added a discussion section, and provided code and data links for reproducibility.
Tissue graph counterfactuals ask how a cell's expression would change under altered spatial neighbor contexts. Such queries are central to predicting cell behavior in tissues, but lack a unified definition, with existing methods targeting specific intervention types or treating cells as i.i.d. In this work, we first formalize tissue graph counterfactuals as a class of spatial interventions that either rewire connections between cells (edge perturbation) or modify the expression of their neighbors (node perturbation). We then introduce Cellina (https://cellina.readthedocs.io) - a framework that uses supervised disentanglement to decompose a cell's intrinsic state from its spatial context, using the latter as a conditioning input for counterfactual predictions. Across benchmarks spanning over 2.5 million spatially-resolved cells in colorectal cancer and mouse brain, Cellina outperforms spatially-informed and non-spatial competitors in in-silico graph perturbations, disentanglement, and scalability. Additionally, we show that Cellina reveals biologically distinct cancer subdomains in an unsupervised manner and enables targeted neighbor perturbation simulations.
Cancer progression involves the sequential accumulation of genetic alterations that cumulatively shape the tumour phenotype. In prostate cancer, tumours can follow divergent evolutionary trajectories that lead to distinct subtypes, but the causes of this divergence remain unclear. While causal inference could elucidate the factors involved, conventional methods are unsuitable due to the possibility of unobserved confounders and ambiguity in the direction of causality. Here, we propose a method that circumvents these issues and apply it to genomic data from 829 prostate cancer patients. We identify several genetic alterations that drive divergence as well as others that prevent this transition, locking tumours into one trajectory. Further analysis reveals that these genetic alterations may cause each other, implying a positive-feedback loop that accelerates divergence. Our findings provide insights into how cancer subtypes emerge and offer a foundation for genomic surveillance strategies aimed at monitoring the progression of prostate cancer.
Motivated by an application in computational biology, we consider low-rank matrix factorization with $\{0,1\}$-constraints on one of the factors and optionally convex constraints on the second one. In addition to the non-convexity shared with other matrix factorization schemes, our problem is further complicated by a combinatorial constraint set of size $2^{m \cdot r}$, where $m$ is the dimension of the data points and $r$ the rank of the factorization. Despite apparent intractability, we provide - in the line of recent work on non-negative matrix factorization by Arora et al. (2012) - an algorithm that provably recovers the underlying factorization in the exact case with $O(m r 2^r + mnr + r^2 n)$ operations for $n$ datapoints. To obtain this result, we use theory around the Littlewood-Offord lemma from combinatorics.