Integrating sequence and gene expression information predicts genome-wide DNA-binding proteins and suggests a cooperative mechanism

Abstract

DNA-binding proteins (DBPs) perform diverse biological functions ranging from transcription to pathogen sensing. Machine learning methods can not only identify DBPs de novo but also provide insights into their DNA-recognition dynamics. However, it remains unclear whether available methods that can accurately predict DNA-binding sites in known DBPs can also identify novel DBPs. Moreover, sequence information is blind to the cellular- and disease-specific contexts of DBP activities, whereas the under-utilized knowledge from public gene expression data offers great promise. To address these issues, we have developed novel methods for predicting DBPs by integrating sequence and gene expression-derived features and applied them to explore human, mouse and Arabidopsis proteomes. While our sequence-based models outperformed the gene expression-based ones, some proteins with weaker DBP-like sequence features were correctly predicted by gene expression-based features, suggesting that these proteins acquire a tangible DBP functionality in a conducive gene expression environment. Analysis of motif enrichment among the co-expressed genes of top 100 candidates DBPs from hitherto unannotated genes provides further avenues to explore their functional associations.

Enabling full‐length evolutionary profiles based deep convolutional neural network for predicting DNA‐binding proteins from sequence

“…DBPs and their functional annotations in this work are taken from our recent work, where Gentle integration of gene expression and sequence attributes (GIGEASA), a model to integrate gene expression and sequence information was reported . In that article, we studied the integration of gene expression and sequence information for DBP prediction for human , mouse , and Arabidopsis Thaliana .…”

“…In this work, we have used the most lenient of these DBP annotation, that is, based on GO term labels. These GO annotations were also collected from GIGEASA . Despite this lenient annotation, the ratio of the number of positive (binding proteins) vs negative class (nonbinding proteins) ranges from 1:30 to 1:10 under various training cycles (see below).…”

“…DBPs and their functional annotations in this work are taken from our recent work, where Gentle integration of gene expression and sequence attributes (GIGEASA), a model to integrate gene expression and sequence information was reported. 31 In that article, we studied F I G U R E 1 Steps involved in DNA-binding protein (DBP) prediction using their zero-padded evolutionary profiles (pPSSM) [Color figure can be viewed at wileyonlinelibrary.com] the integration of gene expression and sequence information for DBP prediction for human, mouse, and Arabidopsis Thaliana. Of these, in the current work, data set from human proteins has been considered, as the results are not expected to be different in other organisms and the objective here is to establish a comparative framework by using a well-studied system as a reference.…”

“…Main summary features traditionally used for function prediction have been (a) amino acid composition of the sequence (see eg, ), (b) dinucleotide or higher k‐mer frequencies, (c) column averages of a position‐specific substitution matrices, representing evolutionary profile or submatrices from a PSSM collected by pooling together rows where the same amino acid residue has been observed . Recently, we extended this feature set by defining sequence features in terms of predicted scores of binding with other ligands' binding sites for each residue in a protein . All these approaches work well and rely upon discarding or summarizing detailed sequence or PSSM information of query proteins, which may be helpful in identifying novel DBPs with greater sensitivity.…”

“…[27][28][29][30] Recently, we extended this feature set by defining sequence features in terms of predicted scores of binding with other ligands' binding sites for each residue in a protein. 31 All these approaches work well and rely upon discarding or summarizing detailed sequence or PSSM information of query proteins, which may be helpful in identifying novel DBPs with greater sensitivity.…”