Biology subjects

Marchal, K.

Publications and source records attributed to Marchal, K..

6 recordsLinked to original sources

bioRxiv2026

pLM-Guided Inverse Folding for Antibody Sequence Design

Inverse folding, predicting amino acid sequences from three-dimensional structures, is a foundational task in computational protein design, yet it is hindered by the scarcity of structural data, which limits model training and risks overfitting. The standard approach fine-tunes general inverse folding models on domain-specific structural datasets like antibodies, but such data remain expensive. To enable inverse folders to benefit more from abundant sequence data, we propose combining ProteinMPNN, a general protein inverse folding model, with IgLM, an antibody-specific language model, via a training-free weighted ensemble of their predictions at inference time. Evaluated on antibody and nanobody structures, our results show that this approach substantially improves amino acid recovery over ProteinMPNN alone, approaching the performance of antibody-specific models like AntiFold while generating more diverse sequences. Even models already fine-tuned on antibody structures (AbMPNN) benefit from language model guidance, demonstrating that it complements structural fine-tuning and leads to more natural-looking sequences that still satisfy structural constraints.

bioinformatics↗

bioRxiv2026

A Conditional Random Field approach for de novo reconstruction of bacterial haplotypes from a de Bruijn graph representation

BackgroundDetecting distinct bacterial strains in a mixed sample is an important, yet less well-developed aspect of metagenomic research. Several methods exist that successfully retrieve a de novo reconstruction of viral strains. However, the reconstruction of bacterial haplotypes poses its own distinct challenges, and methods that successfully reconstruct full genome-length bacterial strains de novo are scarce. Here, we develop HaploDetox, a method for de novo bacterial haplotype reconstruction from short reads. We use a de Bruijn graph representation of the reads in which nodes correspond with k-mers from the read set and arcs represent overlap between two nodes sequences. Our aim is to accurately assign labels to each node and arc in the graph to reveal the presence or absence of their corresponding sequence in individual strains. ResultsUsing a negative binomial mixture model, we model the relationship between the read coverage of nodes and arcs in the graph and their presence in a strain. We achieve improved labelling accuracy by including contextual information from neighbouring nodes and arcs with a Conditional Random Field. These labels are used to extract strain-specific de Bruijn graphs from the original graph. Additionally, we allow users to assess the number of strains present in the dataset based on model selection criteria. We evaluate our node/arc labelling accuracy on simulated datasets and in silico mixes of real datasets containing different numbers of strains, as well as on in vitro mixed real datasets. Existing de novo haplotype reconstruction methods present their reconstruction as strain-specific sets of SNPs. We demonstrate that HaploDetox assigns strain-specific SNPs with a higher recall and similar precision than existing methods, by aligning the unitigs from strain-specific graphs to a reference genome. ConclusionsWe achieve improved strain-specific SNP phasing accuracy as compared to existing methods for de novo bacterial haplotype reconstruction. Additionally, HaploDetox is not limited to the determination of strain-specific SNPs, and other types of variant calls can be obtained through reference alignment. Finally, strain-specific de Bruijn graphs are an important first step towards full genome-length bacterial haplotype-aware assembly.

bioinformatics↗

bioRxiv2023

Digital profiling of cancer transcriptomes from histology images with grouped vision attention

Cancer is a heterogeneous disease that demands precise molecular profiling for better understanding and management. Recently, deep learning has demonstrated potentials for cost-efficient prediction of molecular alterations from histology images. While transformer-based deep learning architectures have enabled significant progress in non-medical domains, their application to histology images remains limited due to small dataset sizes coupled with the explosion of trainable parameters. Here, we develop SEQUOIA, a transformer model to predict cancer transcriptomes from whole-slide histology images. To enable the full potential of transformers, we first pre-train the model using data from 1,802 normal tissues. Then, we fine-tune and evaluate the model in 4,331 tumor samples across nine cancer types. The prediction performance is assessed at individual gene levels and pathway levels through Pearson correlation analysis and root mean square error. The generalization capacity is validated across two independent cohorts comprising 1,305 tumors. In predicting the expression levels of 25,749 genes, the highest performance is observed in cancers from breast, kidney and lung, where SEQUOIA accurately predicts the expression of 11,069, 10,086 and 8,759 genes, respectively. The accurately predicted genes are associated with the regulation of inflammatory response, cell cycles and metabolisms. While the model is trained at the tissue level, we showcase its potential in predicting spatial gene expression patterns using spatial transcriptomics datasets. Leveraging the prediction performance, we develop a digital gene expression signature that predicts the risk of recurrence in breast cancer. SEQUOIA deciphers clinically relevant gene expression patterns from histology images, opening avenues for improved cancer management and personalized therapies.

bioinformatics↗

bioRxiv2022

Benchmark of tools for in silico prediction of MHC class I and class II genotypes from NGS data

The Human Leukocyte Antigen (HLA) genes are a group of highly polymorphic genes that are located in the Major Histocompatibility Complex (MHC) region on chromosome 6. The HLA genotype affects the presentability of tumour antigens to the immune system. While knowledge of these genotypes is of utmost importance to study differences in immune responses between cancer patients, gold standard, PCR-derived genotypes are rarely available in large Next Generation Sequencing (NGS) datasets. Therefore, a variety of methods for in silico NGS-based HLA genotyping have been developed, bypassing the need to determine these genotypes with separate experiments. However, there is currently no consensus on the best performing tool. Here, we compiled a list of 13 HLA callers and evaluated their accuracy on three different datasets. Based on these results, best-practice guidelines were constructed, and consensus HLA allele predictions were made for DNA and RNA samples from The Cancer Genome Atlas (TCGA).

bioinformatics↗

bioRxiv2020

DNA-SIP and repeated isolation corroborate Variovorax as a key organism in maintaining the genetic memory for linuron biodegradation in an agricultural soil

The frequent exposure of agricultural soils to pesticides often leads to microbial adaptation, including the development of dedicated microbial populations that utilize the pesticide compound as a carbon and energy source. Soil from an agricultural field in Halen (Belgium) with a history of linuron exposure has been studied for its linuron-degrading bacterial populations at two time points over the past decade and Variovorax was appointed as a key linuron degrader. Like most studies on pesticide degradation, these studies relied on isolates that were retrieved through bias-prone enrichment procedures and therefore might not represent the in situ active pesticide-degrading populations. In this study, we revisited the Halen field and applied, in addition to enrichment-based isolation, DNA stable isotope probing (DNA-SIP), to identify the in situ linuron degrading bacteria. DNA-SIP unambiguously linked Variovorax and its linuron catabolic genes to linuron dissipation, likely through synergistic cooperation between two species. Additionally, two linuron mineralizing Variovorax isolates were obtained with high 16S rRNA gene sequence similarity to strains isolated from the same field a decade earlier. The results confirm Variovorax as the in situ degrader of linuron in the studied agricultural field and corroborate the genus as key in the maintenance of a robust genetic memory regarding linuron degradation functionality in the examined field.

microbiology↗

bioRxiv2018

Pathway and network analysis of more than 2,500 whole cancer genomes

The catalog of cancer driver mutations in protein-coding genes has greatly expanded in the past decade. However, non-coding cancer driver mutations are less well-characterized and only a handful of recurrent non-coding mutations, most notably TERT promoter mutations, have been reported. Motivated by the success of pathway and network analyses in prioritizing rare mutations in protein-coding genes, we performed multi-faceted pathway and network analyses of non-coding mutations across 2,583 whole cancer genomes from 27 tumor types compiled by the ICGC/TCGA PCAWG project. While few non-coding genomic elements were recurrently mutated in this cohort, we identified 93 genes harboring non-coding mutations that cluster into several modules of interacting proteins. Among these are promoter mutations associated with reduced mRNA expression in TP53, TLE4, and TCF4. We found that biological processes had variable proportions of coding and non-coding mutations, with chromatin remodeling and proliferation pathways altered primarily by coding mutations, while developmental pathways, including Wnt and Notch, altered by both coding and non-coding mutations. RNA splicing was primarily targeted by non-coding mutations in this cohort, with samples containing non-coding mutations exhibiting similar gene expression signatures as coding mutations in well-known RNA splicing factors. These analyses contribute a new repertoire of possible cancer genes and mechanisms that are altered by non-coding mutations and offer insights into additional cancer vulnerabilities that can be investigated for potential therapeutic treatments.

cancer biology↗