2025 Hyper Recent •CC0 1.0 Universal

This work is dedicated to the public domain. No rights reserved.

Access Preprint From Server
June 30th, 2025
Version: 1
The Hong Kong University of Science and Technology
bioinformatics
biorxiv

Genomic Touchstone: Benchmarking Genomic Language Models in the Context of the Central Dogma

Wang, Y.Open in Google Scholar•Cai, Z.Open in Google Scholar•Zeng, Q.Open in Google Scholar•Gao, Y.Open in Google Scholar•Ouyang, J.Open in Google Scholar•Xu, Y.Open in Google Scholar•Yang, S.Open in Google Scholar•He, S.Open in Google Scholar•Nie, Y.Open in Google Scholar•Cai, Y.Open in Google Scholaret al.

The emergence of genomic language models (gLMs) has revolutionized the analysis of genomic sequences, enabling robust capture of biologically meaningful patterns from DNA sequences for an improved understanding of human genome-wide regulatory programs, variant pathogenicity and therapeutic discovery. Given that DNA serves as the foundational blueprint within the central dogma, the ultimate evaluation of a gLM is its ability to generalize across this entire biological cascade. However, existing evaluations lack this holistic and consistent framework, leaving researchers uncertain about which model best translates sequence understanding into downstream biological prediction. Here we present Genomic Touchstone, a comprehensive benchmark designed to evaluate gLMs across 36 diverse tasks and 88 datasets structured along the central dogma\'s modalities of DNA, RNA, and protein, encompassing 5.34 billion base pairs of genomic sequences. We evaluate 34 representative models encompassing transformers and convolutional neural networks, as well as emerging efficient architectures such as Hyena and Mamba. Our analysis yield four key insights. First, gLMs achieve comparable or superior performance on RNA and protein tasks compared to models pretrained on these molecules. Second, transformer-based models continue to lead in overall performance, yet efficient sequence models show promising task-specific capabilities and deserve further exploration. Third, the scaling behavior of gLMs remains incompletely understood. While longer input sequences and more diverse pretraining data consistently improve performance, increases in model size do not always translate into better results. Fourth, pretraining strategies, including the choice of training objectives and the composition of pretraining corpora, exert substantial influence on downstream generalization across different genomic contexts. Genomic Touchstone establishes a unified evaluation framework tailored to human genomics. By spanning multiple molecular modalities and biological tasks, it offers a valuable foundation for guiding future gLM design and understanding model generalizability in complex biological contexts.

Similar Papers

biorxiv
Mon Jun 30 2025
Quantitative analysis of genetic interactions in human cells from genome-wide CRISPR-Cas9 screens
Genetic interaction (GI) networks in model organisms have revealed how combinations of genome variants can impact phenotypes and underscored the value of GI maps for functional genomics. To advance efforts toward a reference human GI network, we developed the quantitative Genetic Interaction (qGI) score, a method for precise GI measurement from genome-wide CRISPR-Cas9 screens in isogenic human cel...
Billmann, M.
•
Costanzo, M.
•
Rahman, M.
•
Chan, K.
...•
Myers, C. L.
biorxiv
Mon Jun 30 2025
STORIES: learning cell fate landscapes from spatial transcriptomics
In dynamic biological processes such as development, spatial transcriptomics is revolutionizing the study of the mechanisms underlying spatial organization within tissues. Inferring cell fate trajectories from spatial transcriptomics profiled at several time points has thus emerged as a critical goal, requiring novel computational methods. Wasserstein gradient flow learning is a promising framewor...
Huizing, G.-J.
•
Samaran, J.
•
Capocefalo, D.
•
Audit, A.
...•
Cantini, L.
biorxiv
Mon Jun 30 2025
Representation Learning Methods for Single-Cell Microscopy are Confounded by Background Cells
Deep learning models are widely used to extract feature representations from microscopy images. While these models are used for single-cell analyses, such as studying single-cell heterogeneity, they typically operate on image crops centered on individual cells with background information present, such as other cells, and it remains unclear to what extent the conclusions of single-cell analyses may...
Gupta, A.
•
Moses, A.
•
Lu, A. X.
biorxiv
Mon Jun 30 2025
FEDRANN: effective long-read overlap detection based on dimensionality reduction and approximate nearest neighbors
Overlap detection is a key step in de novo genome assembly pipelines based on the Overlap-Layout-Consensus (OLC) paradigm. However, existing methods for overlap detection either rely on heuristic seed-and-extension strategies or locality-sensitive hashing (LSH), both of which struggle to handle repetitive genomic regions and the computational burden of large-scale datasets. Here, we present FEDRAN...
Zhang, J.-Y.
•
Miao, C.
•
Qiu, T.
•
Xia, X.
...•
Dong, Y.
biorxiv
Mon Jun 30 2025
Molecular characterization of unique multi-domain harbouring fungal rhodopsin for establishing their novel opto-synthetic biological usages
Organisms employ light as an external stimulus for regulating cellular functions. The light-sensitive photoreceptors detect light at varying wavelengths, activating signaling cascades and triggering a range of physiological responses. Rhodopsin is a transmembrane heptahelical protein that functions as an ion channel, or a pump, and sensory receptor, respectively. It consists of a light-sensing chr...
Kumari, A.
•
Kumar, A.
•
Sharma, K.
•
Pati, S. R.
...•
KATERIYA, S.
biorxiv
Mon Jun 30 2025
A Systematic Benchmark of High-Accuracy PacBio Long-Read RNA Sequencing for Transcript-Level Quantification
PacBio long-read RNA sequencing resolves transcripts with greater clarity than short-read technologies, yet its quantitative performance remains under-evaluated at scale. Here, we benchmark the high-throughput PacBio Kinnex platform against Illumina short-read RNA-seq using matched, deeply sequenced datasets across a time course of endothelial cell differentiation. Compared to Illumina, Kinnex ach...
Wissel, D.
•
Mehlferber, M. M.
•
Nguyen, K. M.
•
Pavelko, V.
...•
Sheynkman, G. M.
biorxiv
Mon Jun 30 2025
scHDeepInsight: A Hierarchical Deep Learning Framework for Precise Immune Cell Annotation in Single-Cell RNA-seq Data
Immune cell classification from single-cell RNA sequencing (scRNA-seq) presents significant challenges due to complex hierarchical relationships among cell types. We introduce scHDeepInsight, a deep learning framework that extends our previous scDeepInsight model by integrating a biologically-informed classification architecture with an adaptive hierarchical focal loss. The framework leverages our...
JIA, S.
•
Lysenko, A.
•
Boroevich, K. A.
•
Sharma, A.
•
Tsunoda, T.
biorxiv
Mon Jun 30 2025
reconcILS: A gene tree-species tree reconciliation algorithm that allows for incomplete lineage sorting
Reconciliation algorithms provide an accounting of the evolutionary history of individual gene trees given a species tree. Many reconciliation algorithms consider only duplication and loss events (and sometimes horizontal transfer), ignoring effects of the coalescent process, including incomplete lineage sorting (ILS). Here, we present a new algorithm for carrying out reconciliation that accuratel...
Mishra, S.
•
Smith, M. L.
•
Hahn, M. W.
biorxiv
Mon Jun 30 2025
Identifying Optimal Machine Learning Approaches for Microbiome-Metabolomics Integration with Stable Feature Selection
Microbiome research has been limited by methodological inconsistencies. Taxonomy-based profiling presents challenges such as data sparsity, variable taxonomic resolution, and the reliance on DNA-based profiling, which provides limited functional insight. Multi-omics integration has emerged as a promising approach to link microbiome composition with function. However, the lack of standardized metho...
Palmer, S. N.
•
Mishra, A. A.
•
Gan, S.
•
Liu, D.
...•
Zhan, X.
biorxiv
Mon Jun 30 2025
CpGeneAge: multi-omics aging clocks associated with Nf-κB signaling pathway in aging
Aging clocks have emerged as the primary tools for measuring biological aging and have been developed for a wide range of single-omic measurements. Epigenetic aging clocks showed high accuracy in age prediction, however, their biological interpretation is still a challenging task. Transcriptomics aging clocks provide better interpretability but worse age prediction accuracy. To exploit the benefit...
Varga, B.
•
Kerepesi, C.