Understanding cellular responses to stimuli is crucial for biological discovery and drug development. Transcriptomics provides interpretable, gene-level insights, while microscopy imaging offers rich predictive features but is harder to interpret. Weakly paired datasets, where samples share biological states, enable multimodal learning but are scarce, limiting their utility for training and multimodal inference. We propose a framework to enhance transcriptomics by distilling knowledge from microscopy images. Using weakly paired data, our method aligns and binds modalities, enriching gene expression representations with morphological information. To address data scarcity, we introduce (1) Semi-Clipped, an adaptation of CLIP for cross-modal distillation using pretrained foundation models, achieving state-of-the-art results, and (2) PEA (Perturbation Embedding Augmentation), a novel augmentation technique that enhances transcriptomics data while preserving inherent biological information. These strategies improve the predictive power and retain the interpretability of transcriptomics, enabling rich unimodal representations for complex biological tasks.

MOTIVATION: Codon reassignments have been reported across all domains of life. With the increasing number of sequenced genomes, the development of systematic approaches for genetic code detection is essential for accurate downstream analyses. Three automated prediction tools exist so far: FACIL, GenDecoder and Bagheera; the last two respectively restricted to metazoan mitochondrial genomes and CUG reassignments in yeast nuclear genomes. These tools can only analyze a single genome at a time and are often not followed by a validation procedure, resulting in a high rate of false positives. RESULTS: We present CoreTracker, a new algorithm for the inference of sense-to-sense codon reassignments. CoreTracker identifies potential codon reassignments in a set of related genomes, then uses statistical evaluations and a random forest classifier to predict those that are the most likely to be correct. Predicted reassignments are then validated through a phylogeny-aware step that evaluates the impact of the new genetic code on the protein alignment. Handling simultaneously a set of genomes in a phylogenetic framework, allows tracing back the evolution of each reassignment, which provides information on its underlying mechanism. Applied to metazoan and yeast genomes, CoreTracker significantly outperforms existing methods on both precision and sensitivity. AVAILABILITY AND IMPLEMENTATION: CoreTracker is written in Python and available at https://github.com/UdeM-LBIT/CoreTracker. CONTACT: [email protected]. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Polytomies in gene trees are multifurcated nodes corresponding to unresolved parts of the tree, usually due to insufficient differentiation between sequences of homologous gene copies. Apart from gene sequences, other information such as that contained in the species tree can be used to resolve such intricate parts of a gene tree. The problem of resolving a multifurcated tree has been considered by many authors, the objective function often being the number of duplications and losses reflected by the reconciliation of the resolved gene tree with the species tree. Here, we present PolytomySolver, an algorithm accounting for a more general model allowing different costs for duplications and losses per species. The time complexity of this algorithm is linear for the unit cost and is quadratic for the general cost, which outperforms the best known solutions so far by a linear factor. We show on simulated trees that the gain in theoretical complexity has a real practical impact on running times.