Biology subjects

Blum, C.

Publications and source records attributed to Blum, C..

7 recordsLinked to original sources

bioRxiv2024

ChatNT: A Multimodal Conversational Agent for DNA, RNA and Protein Tasks

Language models are thriving, powering conversational agents that assist and empower humans to solve a number of tasks. Recently, these models were extended to support additional modalities including vision, audio and video, demonstrating impressive capabilities across multiple domains including healthcare. Still, conversational agents remain limited in biology as they cannot yet fully comprehend biological sequences. On the other hand, high-performance foundation models for biological sequences have been built through self-supervision over sequencing data, but these need to be fine-tuned for each specific application, preventing transfer and generalization between tasks. In addition, these models are not conversational which limits their utility to users with coding capabilities. In this paper, we propose to bridge the gap between biology foundation models and conversational agents by introducing ChatNT, the first multimodal conversational agent with an advanced understanding of biological sequences. ChatNT achieves new state-of-the-art results on the Nucleotide Transformer benchmark while being able to solve all tasks at once, in English, and to generalize to unseen questions. In addition, we have curated a new set of more biologically relevant instructions tasks from DNA, RNA and proteins, spanning multiple species, tissues and biological processes. ChatNT reaches performance on par with state-of-the-art specialized methods on those tasks. We also present a novel perplexity-based technique to help calibrate the confidence of our model predictions. Our framework for genomics instruction-tuning can be easily extended to more tasks and biological data modalities (e.g. structure, imaging), making it a widely applicable tool for biology. ChatNT is the first model of its kind and constitutes an initial step towards building generally capable agents that understand biology from first principles while being accessible to users with no coding background.

genomics↗

bioRxiv2024

SegmentNT: annotating the genome at single-nucleotide resolution with DNA foundation models

Genome annotation models that directly analyze DNA sequences are indispensable for modern biological research, enabling rapid and accurate identification of genes and other functional elements. This capability is paramount as the volume of sequenced genomes rapidly expands, making the need for efficient and accurate annotation methods increasingly critical, particularly in the context of genetic variant prediction and in-silico sequence design. Current annotation tools are typically developed for specific element classes and trained from scratch using supervised learning on datasets that are often limited in size. This approach constrains their performance and ability to generalize to new genomes. Here, we frame the genome annotation problem as instance segmentation and introduce a novel methodology for fine-tuning pre-trained DNA foundation models to segment 14 different genic and regulatory elements at single-nucleotide resolution. We leverage the self-supervised pre-trained model Nucleotide Transformer (NT) to develop a general segmentation model, SegmentNT, capable of processing DNA sequences up to 50kb long. By utilizing pre-trained weights from NT, SegmentNT surpasses the performance of several ablation models and baselines, including convolutional networks with one-hot encoded nucleotide sequences and large models trained from scratch. We demonstrate state-of-the-art performance on gene annotation, splice site and regulatory elements detection throughout the genome. We also leveraged our framework to accommodate two extra DNA foundation models, Enformer and Borzoi, extending the sequence context up to 500kb and enhancing performance on regulatory elements. Finally, we show that a SegmentNT model trained on human genomic elements generalizes to elements of different species, and a multi-species SegmentNT model achieves strong generalization across unseen species. Our approach is readily extensible to additional genomic elements and species. We have made our SegmentNT human and multi-species models, as well as the SegmentEnformer and SegmentBorzoi models, available on our github repository in Jax and HuggingFace space in Pytorch.

genomics↗

bioRxiv2023

A minimal physical model for curvotaxis driven by curved protein complexes at the cell's leading edge

Cells often migrate on curved surfaces inside the body, such as curved tissues, blood vessels or highly curved protrusions of other cells. Recent in-vitro experiments provide clear evidence that motile cells are affected by the curvature of the substrate on which they migrate, preferring certain curvatures to others, termed "curvotaxis". The origin and underlying mechanism that gives rise to this curvature sensitivity are not well understood. Here, we employ a "minimal cell" model which is composed of a vesicle that contains curved membrane protein complexes, that exert protrusive forces on the membrane (representing the pressure due to actin polymerization). This minimal-cell model gives rise to spontaneous emergence of a motile phenotype, driven by a lamellipodia-like leading edge. By systematically screening the behaviour of this model on different types of curved substrates (sinusoidal, cylinder and tube), we show that minimal ingredients and energy terms capture the experimental data. The model recovers the observed migration on the sinusoidal substrate, where cells move along the grooves (minima), while avoiding motion along the ridges. In addition, the model predicts the tendency of cells to migrate circumferentially on convex substrates and axially on concave ones. Both of these predictions are verified experimentally, on several cell types. Altogether, our results identify the minimization of membrane-substrate adhesion energy and binding energy between the membrane protein complexes as key players of curvotaxis in cell migration.

biophysics↗

bioRxiv2023

SARS-CoV-2 N-protein induces the formation of composite α-synuclein/N-protein fibrils that transform into a strain of α-synuclein fibrils

The presence of deposits of alpha-synuclein fibrils in cells of the brain are a hallmark of several -synucleinopathies, including Parkinsons disease. As most disease cases are not familial, it is likely that external factors play a role in disease onset. One of the external factors that may influence disease onset are viral infections. It has recently been shown that in the presence of SARS-Cov-2 N-protein, S fibril formation is faster and proceeds in an unusual two-step aggregation process. Here, we show that faster fibril formation is not due to a SARS-CoV-2 N-protein-catalysed formation of an aggregation-prone nucleus. Instead, aggregation starts with the formation of a population of mixed S/N-protein fibrils with low affinity for S. After the depletion of N-protein, fibril formation comes to a halt, until a slow transformation to fibrils with characteristics of pure S fibril strains occurs. This transformation into a strain of S fibrils subsequently results in a second phase of fibril growth until a new equilibrium is reached. Our findings point at the possible relevance of fibril strain transformation in the cell-to-cell spread of the S pathology and disease onset.

biophysics↗

bioRxiv2021

An accelerated thrombosis model for computational fluid dynamics simulations in rotary blood pumps

PurposeThrombosis is one of the major complications in blood-carrying medical devices and a better understanding to influence design of such devices is desirable. Over the past years many computational models of thrombosis have been developed. However, open questions remain about the applicability and implementation within a pump development process. The aim of the study was to develop and test a computationally efficient model for thrombus risk prediction in rotary blood pumps. MethodsWe used a two-stage approach to calculate thrombus risk. At the first stage, the velocity and pressure fields were computed by computational fluid dynamic (CFD) simulations. At the second stage, platelet activation by mechanical and chemical stimuli was determined through species transport with an Eulerian approach. The model was implemented in ANSYS CFX and compared with existing clinical data on thrombus deposition within the HeartMate II. ResultsOur model shows good correlation (R2>0.94) with clinical data and identifies the bearing and outlet stator region of the HeartMate II as the location most prone to thrombus formation. The calculation of platelet activation requires an additional 10-20 core hours of computation time. DiscussionThe concentration of activated platelets can be used as a surrogate marker to determine risk regions of thrombus deposition in a blood pump. Model expansion, e.g. by including more chemical species can easily be performed. We make our model openly available by implementing it for the FDA benchmark blood pump. DeclarationsO_ST_ABSFundingC_ST_ABSThis research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. Open access funding enabled and organized by Projekt DEAL. Conflict of interestAll of the authors have nothing to disclose. Availability of data and materialThe raw data can be retrieved by request from the authors. Code availabilityThe implementation of the thrombus model in the FDA benchmark blood pump geometry is available on https://doi.org/10.5281/zenodo.5116063. Authors contributionsAll authors contributed to the study conception and design. CB developed the numerical model, performed the simulations, gathered, analysed and discussed the results. SGH, MN and US were involved in the analysis and discussion of the results. MN supervised the project. MN and CB wrote the manuscript based on the input of all co-authors. All co-authors read and approved the final version of the manuscript.

bioengineering↗

bioRxiv2021

Interactions between SARS-CoV-2 N-protein and α-synuclein accelerate amyloid formation

First cases that point at a correlation between SARS-CoV-2 infections and the development of Parkinsons disease have been reported. Currently it is unclear if there also is a direct causal link between these diseases. To obtain first insights into a possible molecular relation between viral infections and the aggregation of -synuclein protein into amyloid fibrils characteristic for Parkinsons disease, we investigated the effect of the presence of SARS-CoV-2 proteins on -synuclein aggregation. We show, in test tube experiments, that SARS-CoV-2 S-protein has no effect on -synuclein aggregation while SARS-CoV-2 N-protein considerably speeds up the aggregation process. We observe the formation of multi-protein complexes, and eventually amyloid fibrils. Microinjection of N-protein in SHSY-5Y cells disturbed the -synuclein proteostasis and increased cell death. Our results point toward direct interactions between the N-protein of SARS-CoV-2 and -synuclein as molecular basis for the observed coincidence between SARS-CoV-2 infections and Parkinsonism.

biophysics↗

bioRxiv2016

Predicting gene expression level in E. coli from mRNA sequence information

MotivationThe accurate characterization of the translational mechanism is crucial for enhancing our understanding of the relationship between genotype and phenotype. In particular, predicting the impact of the genetic variants on gene expression will allow to optimize specific pathways and functions for engineering new biological systems. In this context, the development of accurate methods for predicting translation efficiency from the nucleotide sequence is a key challenge in computational biology.\n\nMethodsIn this work we present PGExpress, a binary classifier to discriminate between mRNA sequences with low and high translation efficiency in E. coli. PGExpress algorithm takes as input 12 features corresponding to RNA folding and anti-Shine-Dalgarno hybridization free energies. The method was trained on a set of 1,772 sequence variants (WT-High) of 137 essential E. coli genes. For each gene, we considered 13 sequence variants of the first 33 nucleotides encoding for the same amino acids followed by the superfolder GFP. Each gene variant is represented sequence blocks that include the Ribosome Binding Site (RBS), the first 33 nucleotides of the coding region (C33), the remaining part of the coding region (CC), and their combinations.\n\nResultsOur logistic regression-based tool (PGExpress) was trained using a 20-fold gene-based cross-validation procedure on the WT-High dataset. In this test PGExpress achieved an overall accuracy of 74%, a Matthews correlation coefficient 0.49 and an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.81. Tested on 3 sets of sequences with different Ribosome Binding Sites, PGExpress reaches similar AUC. Finally, we validated our method by performing in-house experiments on five newly generated mRNA sequence variants. The predictions of the expression level of the new variants are in agreement with our experimental results in E. coli.\n\nAvailabilityhttp://folding.biofold.org/pgexpress\n\nContactmarkus.kollmann@hhu.de, emidio.capriotti@unibo.it

bioinformatics↗