Integrating convolution and self-attention improves language model of human genome for interpreting non-coding regions at base-resolution

Genomic Language Models: Opportunities and Challenges

Large language models (LLMs) are having transformative impacts across a wide range of scientific fields, particularly in the biomedical sciences. Just as the goal of Natural Language Processing is to understand sequences of words, a major objective in biology is to understand biological sequences. Genomic Language Models (gLMs), which are LLMs trained on DNA sequences, have the potential to significantly advance our understanding of genomes and how DNA elements at various scales interact to give rise to complex functions. To showcase this potential, we highlight key applications of gLMs, including functional constraint prediction, sequence design, and transfer learning. Despite notable recent progress, however, developing effective and efficient gLMs presents numerous challenges, especially for species with large, complex genomes. Here, we discuss major considerations for developing and evaluating gLMs.

Deciphering 3'UTR Mediated Gene Regulation Using Interpretable Deep Representation Learning

The 3' untranslated regions (3'UTRs) of messenger RNAs contain many important cis-regulatory elements that are under functional and evolutionary constraints. It is hypothesized that these constraints are similar to grammars and syntaxes in human languages and can be modeled by advanced natural language techniques such as Transformers, which has been very effective in modeling complex protein sequence and structures. Here 3UTRBERT is described, which implements an attention-based language model, i.e., Bidirectional Encoder Representations from Transformers (BERT). 3UTRBERT is pre-trained on aggregated 3'UTR sequences of human mRNAs in a task-agnostic manner; the pre-trained model is then fine-tuned for specific downstream tasks such as identifying RBP binding sites, m6A RNA modification sites, and predicting RNA sub-cellular localizations. Benchmark results show that 3UTRBERT generally outperformed other contemporary methods in each of these tasks. More importantly, the self-attention mechanism within 3UTRBERT allows direct visualization of the semantic relationship between sequence elements and effectively identifies regions with important regulatory potential. It is expected that 3UTRBERT model can serve as the foundational tool to analyze various sequence labeling tasks within the 3'UTR fields, thus enhancing the decipherability of post-transcriptional regulatory mechanisms.

A foundational large language model for edible plant genomes

Significant progress has been made in the field of plant genomics, as demonstrated by the increased use of high-throughput methodologies that enable the characterization of multiple genome-wide molecular phenotypes. These findings have provided valuable insights into plant traits and their underlying genetic mechanisms, particularly in model plant species. Nonetheless, effectively leveraging them to make accurate predictions represents a critical step in crop genomic improvement. We present AgroNT, a foundational large language model trained on genomes from 48 plant species with a predominant focus on crop species. We show that AgroNT can obtain state-of-the-art predictions for regulatory annotations, promoter/terminator strength, tissue-specific gene expression, and prioritize functional variants. We conduct a large-scale in silico saturation mutagenesis analysis on cassava to evaluate the regulatory impact of over 10 million mutations and provide their predicted effects as a resource for variant characterization. Finally, we propose the use of the diverse datasets compiled here as the Plants Genomic Benchmark (PGB), providing a comprehensive benchmark for deep learning-based methods in plant genomic research. The pre-trained AgroNT model is publicly available on HuggingFace at https://huggingface.co/InstaDeepAI/agro-nucleotide-transformer-1b  for future research purposes.

Self-supervised learning on millions of primary RNA sequences from 72 vertebrates improves sequence-based RNA splicing prediction

Language models pretrained by self-supervised learning (SSL) have been widely utilized to study protein sequences, while few models were developed for genomic sequences and were limited to single species. Due to the lack of genomes from different species, these models cannot effectively leverage evolutionary information. In this study, we have developed SpliceBERT, a language model pretrained on primary ribonucleic acids (RNA) sequences from 72 vertebrates by masked language modeling, and applied it to sequence-based modeling of RNA splicing. Pretraining SpliceBERT on diverse species enables effective identification of evolutionarily conserved elements. Meanwhile, the learned hidden states and attention weights can characterize the biological properties of splice sites. As a result, SpliceBERT was shown effective on several downstream tasks: zero-shot prediction of variant effects on splicing, prediction of branchpoints in humans, and cross-species prediction of splice sites. Our study highlighted the importance of pretraining genomic language models on a diverse range of species and suggested that SSL is a promising approach to enhance our understanding of the regulatory logic underlying genomic sequences.

Evaluating the representational power of pre-trained DNA language models for regulatory genomics

The emergence of genomic language models (gLMs) offers an unsupervised approach to learn a wide diversity of cis-regulatory patterns in the non-coding genome without requiring labels of functional activity generated by wet-lab experiments. Previous evaluations have shown pre-trained gLMs can be leveraged to improve prediction performance across a broad range of regulatory genomics tasks, albeit using relatively simple benchmark datasets and baseline models. Since the gLMs in these studies were tested upon fine-tuning their weights for each downstream task, determining whether gLM representations embody a foundational understanding of cis-regulatory biology remains an open question. Here we evaluate the representational power of pre-trained gLMs to predict and interpret cell-type-specific functional genomics data that span DNA and RNA regulation. Our findings suggest that current gLMs do not offer substantial advantages over conventional machine learning approaches that use one-hot encoded sequences. This work highlights a major limitation with current gLMs, raising potential issues in conventional pre-training strategies for the non-coding genome.

SetQuence & SetOmic: Deep set transformers for whole genome and exome tumour analysis

In oncology, Deep Learning has shown great potential to personalise tasks such as tumour type classification, based on per-patient omics data-sets. Being high dimensional, incorporation of such data in one model is a challenge, often leading to one-dimensional studies and, therefore, information loss. Instead, we first propose relying on non-fixed sets of whole genome or whole exome variant-associated sequences, which can be used for supervised learning of oncology-relevant tasks by our Set Transformer based Deep Neural Network, SetQuence. We optimise this architecture to improve its efficiency. This allows for exploration of not just coding but also non-coding variants, from large datasets. Second, we extend the model to incorporate these representations together with multiple other sources of omics data in a flexible way with SetOmic. Evaluation, using these representations, shows improved robustness and reduced information loss compared to previous approaches, while still being computationally tractable. By means of Explainable Artificial Intelligence methods, our models are able to recapitulate the biological contribution of highly attributed features in the tumours studied. This validation opens the door to novel directions in multi-faceted genome and exome wide biomarker discovery and personalised treatment among other presently clinically relevant tasks.

DNA language models are powerful predictors of genome-wide variant effects

The expanding catalog of genome-wide association studies (GWAS) provides biological insights across a variety of species, but identifying the causal variants behind these associations remains a significant challenge. Experimental validation is both labor-intensive and costly, highlighting the need for accurate, scalable computational methods to predict the effects of genetic variants across the entire genome. Inspired by recent progress in natural language processing, unsupervised pretraining on large protein sequence databases has proven successful in extracting complex information related to proteins. These models showcase their ability to learn variant effects in coding regions using an unsupervised approach. Expanding on this idea, we here introduce the Genomic Pre-trained Network (GPN), a model designed to learn genome-wide variant effects through unsupervised pretraining on genomic DNA sequences. Our model also successfully learns gene structure and DNA motifs without any supervision. To demonstrate its utility, we train GPN on unaligned reference genomes of Arabidopsis thaliana and seven related species within the Brassicales order and evaluate its ability to predict the functional impact of genetic variants in A. thaliana by utilizing allele frequencies from the 1001 Genomes Project and a comprehensive database of GWAS. Notably, GPN outperforms predictors based on popular conservation scores such as phyloP and phastCons. Our predictions for A. thaliana can be visualized as sequence logos in the UCSC Genome Browser (https://genome.ucsc.edu/s/gbenegas/gpn-arabidopsis). We provide code (https://github.com/songlab-cal/gpn) to train GPN for any given species using its DNA sequence alone, enabling unsupervised prediction of variant effects across the entire genome.

An Overview of Deep Generative Models in Functional and Evolutionary Genomics

Following the widespread use of deep learning for genomics, deep generative modeling is also becoming a viable methodology for the broad field. Deep generative models (DGMs) can learn the complex structure of genomic data and allow researchers to generate novel genomic instances that retain the real characteristics of the original dataset. Aside from data generation, DGMs can also be used for dimensionality reduction by mapping the data space to a latent space, as well as for prediction tasks via exploitation of this learned mapping or supervised/semi-supervised DGM designs. In this review, we briefly introduce generative modeling and two currently prevailing architectures, we present conceptual applications along with notable examples in functional and evolutionary genomics, and we provide our perspective on potential challenges and future directions.

MuLan-Methyl-multiple transformer-based language models for accurate DNA methylation prediction

Transformer-based language models are successfully used to address massive text-related tasks. DNA methylation is an important epigenetic mechanism, and its analysis provides valuable insights into gene regulation and biomarker identification. Several deep learning-based methods have been proposed to identify DNA methylation, and each seeks to strike a balance between computational effort and accuracy. Here, we introduce MuLan-Methyl, a deep learning framework for predicting DNA methylation sites, which is based on 5 popular transformer-based language models. The framework identifies methylation sites for 3 different types of DNA methylation: N6-adenine, N4-cytosine, and 5-hydroxymethylcytosine. Each of the employed language models is adapted to the task using the "pretrain and fine-tune" paradigm. Pretraining is performed on a custom corpus of DNA fragments and taxonomy lineages using self-supervised learning. Fine-tuning aims at predicting the DNA methylation status of each type. The 5 models are used to collectively predict the DNA methylation status. We report excellent performance of MuLan-Methyl on a benchmark dataset. Moreover, we argue that the model captures characteristic differences between different species that are relevant for methylation. This work demonstrates that language models can be successfully adapted to applications in biological sequence analysis and that joint utilization of different language models improves model performance. Mulan-Methyl is open source, and we provide a web server that implements the approach.

Uncovering the Relationship between Tissue-Specific TF-DNA Binding and Chromatin Features through a Transformer-Based Model

Chromatin features can reveal tissue-specific TF-DNA binding, which leads to a better understanding of many critical physiological processes. Accurately identifying TF-DNA bindings and constructing their relationships with chromatin features is a long-standing goal in the bioinformatic field. However, this has remained elusive due to the complex binding mechanisms and heterogeneity among inputs. Here, we have developed the GHTNet (General Hybrid Transformer Network), a transformer-based model to predict TF-DNA binding specificity. The GHTNet decodes the relationship between tissue-specific TF-DNA binding and chromatin features via a specific input scheme of alternative inputs and reveals important gene regions and tissue-specific motifs. Our experiments show that the GHTNet has excellent performance, achieving about a 5% absolute improvement over existing methods. The TF-DNA binding mechanism analysis shows that the importance of TF-DNA binding features varies across tissues. The best predictor is based on the DNA sequence, followed by epigenomics and shape. In addition, cross-species studies address the limited data, thus providing new ideas in this case. Moreover, the GHTNet is applied to interpret the relationship among TFs, chromatin features, and diseases associated with AD46 tissue. This paper demonstrates that the GHTNet is an accurate and robust framework for deciphering tissue-specific TF-DNA binding and interpreting non-coding regions.