Learning the Regulatory Code of Gene Expression

Abstract

Data-driven machine learning is the method of choice for predicting molecular phenotypes from nucleotide sequence, modeling gene expression events including protein-DNA binding, chromatin states as well as mRNA and protein levels. Deep neural networks automatically learn informative sequence representations and interpreting them enables us to improve our understanding of the regulatory code governing gene expression. Here, we review the latest developments that apply shallow or deep learning to quantify molecular phenotypes and decode the cis-regulatory grammar from prokaryotic and eukaryotic sequencing data. Our approach is to build from the ground up, first focusing on the initiating protein-DNA interactions, then specific coding and non-coding regions, and finally on advances that combine multiple parts of the gene and mRNA regulatory structures, achieving unprecedented performance. We thus provide a quantitative view of gene expression regulation from nucleotide sequence, concluding with an information-centric overview of the central dogma of molecular biology.

Generative and interpretable machine learning for aptamer design and analysis of in vitro sequence selection

“…We believe our approach has the potential to generate alternative or better binders for these complex targets, as well as to unveil the sequence motifs that are enriched or avoided in these high-quality aptamers. The same approach can be also useful to model RNA and DNA regulatory sequences and their interaction with proteins in the key processes such as transcription regulation [52, 26, 56, 23]. Lastly, our modeling and design methods are also readily applicable to other selection-amplification protocols, such as phage display for antibody discovery [14, 24] or directed protein evolution studies [4, 42], which have much larger space of possible sequences (20 L for length L ) compared to aptamers (4 L ).…”

“…Over the last decade, deep neural networks (DNN) have become a popular machine learning tool in many areas, such as image recognition or natural language processing, and are now increasingly applied in chemical and biological data processing workflow [22, 56, 23, 6]. However, training DNNs typically requires large datasets, which can be challenging and expensive to obtain from biological experiments.…”

Toward learning the principles of plant gene regulation

“…This contributes to the awareness that gene expression regulation spans different coding and noncoding regions that include the enhancer, promoter, untranslated regions (UTRs), and terminator [2,8] (Figure 1A). It is affected by the enzymatic accessibility of DNA defined by chromatin and epigenetic states [3,10]. Since mRNA abundance is a result of both mRNA synthesis and degradation, it is controlled not only by TFs and core promoters, but by a more complex set of cis-acting elements carried mostly by UTRs [11].…”

“…Since mRNA abundance is a result of both mRNA synthesis and degradation, it is controlled not only by TFs and core promoters, but by a more complex set of cis-acting elements carried mostly by UTRs [11]. Whereas promoter regions were found to explain up to 96% of the variation of gene expression according to DNNs, coding regions can explain up to 69% and 5′ and 3′ UTRs as much as 89% [10]. The different regions also carry complementary information, with different parts coevolving and predictive of the activity of others [2].…”

Virtual Gene Concept and a Corresponding Pragmatic Research Program in Genetical Data Science

“…Deep learning is currently making a great impact across all these related fields. Its applications have already been reviewed for general omics [ 126 , 127 , 128 , 129 ], gene function prediction [ 130 , 131 ], disease prediction [ 132 ], predicting the impact of genetic variation in genomics [ 129 , 133 ], predicting gene regulatory networks [ 133 , 134 ], regulatory genomics [ 135 ], sequence motifs of transcription factors and enhancers [ 133 , 134 , 136 , 137 , 138 , 139 , 140 , 141 , 142 ], variant calling and pathogenicity scores [ 143 ], precision medicine [ 144 , 145 ], pharmacogenomics [ 128 ], and even the prediction of CRISPR targets [ 146 ].…”