q-bio.GN

arXiv:2410.22452v2 Announce Type: replace-cross Abstract: Identifying mutations of SARS-CoV-2 strains associated with their phenotypic changes is critical for pandemic prediction and prevention. We compared an explainable convolutional neural network (CNN) approach and the traditional genome-wide association study (GWAS) on the mutations associated with WHO labels of SARS-CoV-2, a proxy for virulence phenotypes. We trained a CNN classification model that can predict genomic sequences into Variants of Concern (VOCs) and then applied Shapley Additive explanations (SHAP) model to identify mutations that are important for the correct predictions. For comparison, we performed traditional GWAS to identify mutations associated with VOCs. Comparison of the two approaches shows that the explainable neural network approach can more effectively reveal known nucleotide substitutions associated with VOCs, such as those in the spike gene regions. Our results suggest that explainable neural networks for genomic sequences offer a promising alternative to the traditional genome wide analysis approaches.

arXiv:2412.18156v1 Announce Type: cross Abstract: Large language models (LLMs) have demonstrated remarkable advancements, primarily due to their capabilities in modeling the hidden relationships within text sequences. This innovation presents a unique opportunity in the field of life sciences, where vast collections of single-cell omics data from multiple species provide a foundation for training foundational models. However, the challenge lies in the disparity of data scales across different species, hindering the development of a comprehensive model for interpreting genetic data across diverse organisms. In this study, we propose an innovative hybrid approach that integrates the general knowledge capabilities of LLMs with domain-specific representation models for single-cell omics data interpretation. We begin by focusing on genes as the fundamental unit of representation. Gene representations are initialized using functional descriptions, leveraging the strengths of mature language models such as LLaMA-2. By inputting single-cell gene-level expression data with prompts, we effectively model cellular representations based on the differential expression levels of genes across various species and cell types. In the experiments, we constructed developmental cells from humans and mice, specifically targeting cells that are challenging to annotate. We evaluated our methodology through basic tasks such as cell annotation and visualization analysis. The results demonstrate the efficacy of our approach compared to other methods using LLMs, highlighting significant improvements in accuracy and interoperability. Our hybrid approach enhances the representation of single-cell data and offers a robust framework for future research in cross-species genetic analysis.

arXiv:2412.18154v1 Announce Type: cross Abstract: Emerging topics in biomedical research are continuously expanding, providing a wealth of information about genes and their function. This rapid proliferation of knowledge presents unprecedented opportunities for scientific discovery and formidable challenges for researchers striving to keep abreast of the latest advancements. One significant challenge is navigating the vast corpus of literature to extract vital gene-related information, a time-consuming and cumbersome task. To enhance the efficiency of this process, it is crucial to address several key challenges: (1) the overwhelming volume of literature, (2) the complexity of gene functions, and (3) the automated integration and generation. In response, we propose GeneSUM, a two-stage automated gene summary extractor utilizing a large language model (LLM). Our approach retrieves and eliminates redundancy of target gene literature and then fine-tunes the LLM to refine and streamline the summarization process. We conducted extensive experiments to validate the efficacy of our proposed framework. The results demonstrate that LLM significantly enhances the integration of gene-specific information, allowing more efficient decision-making in ongoing research.

arXiv:2411.03871v2 Announce Type: replace Abstract: A common step at the core of many RNA transcript assembly tools is to find a set of weighted paths that best explain the weights of a DAG. While such problems easily become NP-hard, scalable solvers exist only for a basic error-free version of this problem, namely minimally decomposing a network flow into weighted paths. The main result of this paper is to show that we can achieve speedups of two orders of magnitude also for path-finding problems in the realistic setting (i.e., the weights do not induce a flow). We obtain these by employing the safety information that is encoded in the graph structure inside Integer Linear Programming (ILP) solvers for these problems. We first characterize the paths that appear in all path covers of the DAG, generalizing a graph reduction commonly used in the error-free setting (e.g. by Kloster et al. [ALENEX~2018]). Secondly, following the work of Ma, Zheng and Kingsford [RECOMB 2021], we characterize the \emph{sequences} of arcs that appear in all path covers of the DAG. We experiment with a path-finding ILP model (least squares) and with a more recent and accurate one. We use a variety of datasets originally created by Shao and Kingsford [TCBB, 2017], as well as graphs built from sequencing reads by the state-of-the-art tool for long-read transcript discovery, IsoQuant [Prjibelski et al., Nat.~Biotechnology~2023]. The ILPs armed with safe paths or sequences exhibit significant speed-ups over the original ones. On graphs with a large width, average speed-ups are in the range $50-160\times$ in the latter ILP model and in the range $100-1000\times$ in the least squares model. Our scaling techniques apply to any ILP whose solution paths are a path cover of the arcs of the DAG. As such, they can become a scalable building block of practical RNA transcript assembly tools, avoiding heuristic trade-offs currently needed on complex graphs.

arXiv:2412.16074v1 Announce Type: new Abstract: DNA data storage is rapidly gaining traction as a long-term data archival solution, primarily due to its exceptional durability. Retrieving stored data relies on DNA sequencing, which involves a process called basecalling -- a typically costly and slow task that uses machine learning to map raw sequencing signals back to individual DNA bases (which are then translated into digital bits to recover the data). Current models for basecalling have been optimized for reading individual bases. However, with the advent of novel DNA synthesis methods tailored for data storage, there is significant potential for optimizing the reading process. In this paper, we focus on Motif-based DNA synthesis, where sequences are constructed from motifs -- groups of bases -- rather than individual bases. To enable efficient reading of data stored in DNA using Motif-based DNA synthesis, we designed Motif Caller, a machine learning model built to detect entire motifs within a DNA sequence, rather than individual bases. Motifs can also be detected from individually identified bases using a basecaller and then searching for motifs, however, such an approach is unnecessarily complex and slow. Building a machine learning model that directly identifies motifs allows to avoid the additional step of searching for motifs. It also makes use of the greater amount of features per motif, thus enabling finding the motifs with higher accuracy. Motif Caller significantly enhances the efficiency and accuracy of data retrieval in DNA storage based on Motif-Based DNA synthesis.

arXiv:2410.19236v2 Announce Type: replace Abstract: With the rapid growth of large-scale machine learning models in genomics, Shapley values have emerged as a popular method for model explanations due to their theoretical guarantees. While Shapley values explain model predictions locally for an individual input query sequence, extracting biological knowledge requires global explanation across thousands of input sequences. This demands exponential model evaluations per sequence, resulting in significant computational cost and carbon footprint. Herein, we develop SHAP zero, a method that estimates Shapley values and interactions with a near-zero marginal cost for future queried sequences after paying a one-time fee for model sketching. SHAP zero achieves this by establishing a surprisingly underexplored connection between the Shapley values and interactions and the Fourier transform of the model. Explaining two genomic models, one trained to predict guide RNA binding and the other to predict DNA repair outcome, we demonstrate that SHAP zero achieves orders of magnitude reduction in amortized computational cost compared to state-of-the-art algorithms, revealing almost all predictive motifs -- a finding previously inaccessible due to the combinatorial space of possible interactions.

arXiv:2406.06969v2 Announce Type: replace-cross Abstract: Background: Understanding cellular diversity throughout the body is essential for elucidating the complex functions of biological systems. Recently, large-scale single-cell omics datasets, known as omics atlases, have become available. These atlases encompass data from diverse tissues and cell-types, providing insights into the landscape of cell-type-specific gene expression. However, the isolated effect of the tissue environment has not been thoroughly investigated. Evaluating this isolated effect is challenging due to statistical confounding with cell-type effects, arising from significant biases in the combinations of tissues and cell-types within the body. Results: This study introduces a novel data analysis framework, named the Combinatorial Sub-dataset Extraction for Confounding Reduction (COSER), which addresses statistical confounding by using graph theory to enumerate appropriate sub-datasets. COSER enables the assessment of isolated effects of discrete variables in single cells. Applying COSER to the Tabula Muris Senis single-cell transcriptome atlas, we characterized the isolated impact of tissue environments. Our findings demonstrate that some of genes are markedly affected by the tissue environment, particularly in modulating intercellular diversity in immune responses and their age-related changes. Conclusion: COSER provides a robust, general-purpose framework for evaluating the isolated effects of discrete variables from large-scale data mining. This approach reveals critical insights into the interplay between tissue environments and gene expression.