{"title":"在调控序列预测方面,使用伪标记的半监督学习与大型语言模型相比效果更佳。","authors":"Han Phan, Céline Brouard, Raphaël Mourad","doi":"10.1093/bib/bbae560","DOIUrl":null,"url":null,"abstract":"<p><p>Predicting molecular processes using deep learning is a promising approach to provide biological insights for non-coding single nucleotide polymorphisms identified in genome-wide association studies. However, most deep learning methods rely on supervised learning, which requires DNA sequences associated with functional data, and whose amount is severely limited by the finite size of the human genome. Conversely, the amount of mammalian DNA sequences is growing exponentially due to ongoing large-scale sequencing projects, but in most cases without functional data. To alleviate the limitations of supervised learning, we propose a novel semi-supervised learning (SSL) based on pseudo-labeling, which allows to exploit unlabeled DNA sequences from numerous genomes during model pre-training. We further improved it incorporating principles from the Noisy Student algorithm to predict the confidence in pseudo-labeled data used for pre-training, which showed improvements for transcription factor with very few binding (very small training data). The approach is very flexible and can be used to train any neural architecture including state-of-the-art models, and shows in most cases strong predictive performance improvements compared to standard supervised learning. Moreover, small models trained by SSL showed similar or better performance than large language model DNABERT2.</p>","PeriodicalId":9209,"journal":{"name":"Briefings in bioinformatics","volume":null,"pages":null},"PeriodicalIF":6.8000,"publicationDate":"2024-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Semi-supervised learning with pseudo-labeling compares favorably with large language models for regulatory sequence prediction.\",\"authors\":\"Han Phan, Céline Brouard, Raphaël Mourad\",\"doi\":\"10.1093/bib/bbae560\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<p><p>Predicting molecular processes using deep learning is a promising approach to provide biological insights for non-coding single nucleotide polymorphisms identified in genome-wide association studies. However, most deep learning methods rely on supervised learning, which requires DNA sequences associated with functional data, and whose amount is severely limited by the finite size of the human genome. Conversely, the amount of mammalian DNA sequences is growing exponentially due to ongoing large-scale sequencing projects, but in most cases without functional data. To alleviate the limitations of supervised learning, we propose a novel semi-supervised learning (SSL) based on pseudo-labeling, which allows to exploit unlabeled DNA sequences from numerous genomes during model pre-training. We further improved it incorporating principles from the Noisy Student algorithm to predict the confidence in pseudo-labeled data used for pre-training, which showed improvements for transcription factor with very few binding (very small training data). The approach is very flexible and can be used to train any neural architecture including state-of-the-art models, and shows in most cases strong predictive performance improvements compared to standard supervised learning. Moreover, small models trained by SSL showed similar or better performance than large language model DNABERT2.</p>\",\"PeriodicalId\":9209,\"journal\":{\"name\":\"Briefings in bioinformatics\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":6.8000,\"publicationDate\":\"2024-09-23\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Briefings in bioinformatics\",\"FirstCategoryId\":\"99\",\"ListUrlMain\":\"https://doi.org/10.1093/bib/bbae560\",\"RegionNum\":2,\"RegionCategory\":\"生物学\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q1\",\"JCRName\":\"BIOCHEMICAL RESEARCH METHODS\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Briefings in bioinformatics","FirstCategoryId":"99","ListUrlMain":"https://doi.org/10.1093/bib/bbae560","RegionNum":2,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"BIOCHEMICAL RESEARCH METHODS","Score":null,"Total":0}
引用次数: 0
摘要
利用深度学习预测分子过程是一种很有前途的方法,可为全基因组关联研究中发现的非编码单核苷酸多态性提供生物学见解。然而,大多数深度学习方法依赖于监督学习,而监督学习需要与功能数据相关的 DNA 序列,其数量受到人类基因组有限大小的严重限制。相反,由于正在进行的大规模测序项目,哺乳动物 DNA 序列的数量呈指数级增长,但在大多数情况下没有功能数据。为了缓解监督学习的局限性,我们提出了一种基于伪标记的新型半监督学习(SSL),可以在模型预训练过程中利用来自众多基因组的未标记 DNA 序列。我们结合噪声学生算法的原理对其进行了进一步改进,以预测用于预训练的伪标记数据的置信度,结果表明,该方法对结合极少(训练数据极少)的转录因子有很大改进。这种方法非常灵活,可用于训练任何神经架构,包括最先进的模型,而且在大多数情况下,与标准监督学习相比,其预测性能都有很大提高。此外,由 SSL 训练的小型模型显示出与大型语言模型 DNABERT2 相似甚至更好的性能。
Semi-supervised learning with pseudo-labeling compares favorably with large language models for regulatory sequence prediction.
Predicting molecular processes using deep learning is a promising approach to provide biological insights for non-coding single nucleotide polymorphisms identified in genome-wide association studies. However, most deep learning methods rely on supervised learning, which requires DNA sequences associated with functional data, and whose amount is severely limited by the finite size of the human genome. Conversely, the amount of mammalian DNA sequences is growing exponentially due to ongoing large-scale sequencing projects, but in most cases without functional data. To alleviate the limitations of supervised learning, we propose a novel semi-supervised learning (SSL) based on pseudo-labeling, which allows to exploit unlabeled DNA sequences from numerous genomes during model pre-training. We further improved it incorporating principles from the Noisy Student algorithm to predict the confidence in pseudo-labeled data used for pre-training, which showed improvements for transcription factor with very few binding (very small training data). The approach is very flexible and can be used to train any neural architecture including state-of-the-art models, and shows in most cases strong predictive performance improvements compared to standard supervised learning. Moreover, small models trained by SSL showed similar or better performance than large language model DNABERT2.
期刊介绍:
Briefings in Bioinformatics is an international journal serving as a platform for researchers and educators in the life sciences. It also appeals to mathematicians, statisticians, and computer scientists applying their expertise to biological challenges. The journal focuses on reviews tailored for users of databases and analytical tools in contemporary genetics, molecular and systems biology. It stands out by offering practical assistance and guidance to non-specialists in computerized methodologies. Covering a wide range from introductory concepts to specific protocols and analyses, the papers address bacterial, plant, fungal, animal, and human data.
The journal's detailed subject areas include genetic studies of phenotypes and genotypes, mapping, DNA sequencing, expression profiling, gene expression studies, microarrays, alignment methods, protein profiles and HMMs, lipids, metabolic and signaling pathways, structure determination and function prediction, phylogenetic studies, and education and training.