Browse
Submit Data
Databases
API
Help

Dataset Information

36 Views

0 Connections

0 Citations

0 Reanalyses

0 Downloads

Omics score: 0

Biologically relevant transfer learning improves transcription factor binding prediction.

ABSTRACT:

Background

Deep learning has proven to be a powerful technique for transcription factor (TF) binding prediction but requires large training datasets. Transfer learning can reduce the amount of data required for deep learning, while improving overall model performance, compared to training a separate model for each new task.

Results

We assess a transfer learning strategy for TF binding prediction consisting of a pre-training step, wherein we train a multi-task model with multiple TFs, and a fine-tuning step, wherein we initialize single-task models for individual TFs with the weights learned by the multi-task model, after which the single-task models are trained at a lower learning rate. We corroborate that transfer learning improves model performance, especially if in the pre-training step the multi-task model is trained with biologically relevant TFs. We show the effectiveness of transfer learning for TFs with ~ 500 ChIP-seq peak regions. Using model interpretation techniques, we demonstrate that the features learned in the pre-training step are refined in the fine-tuning step to resemble the binding motif of the target TF (i.e., the recipient of transfer learning in the fine-tuning step). Moreover, pre-training with biologically relevant TFs allows single-task models in the fine-tuning step to learn useful features other than the motif of the target TF.

Conclusions

Our results confirm that transfer learning is a powerful technique for TF binding prediction.

SUBMITTER: Novakovsky G

PROVIDER: S-EPMC8474956 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Json Xml

Similar Datasets

Prediction of Transcription Factor Binding Sites Using a Combined Deep Learning Approach.

Project description:In the process of regulating gene expression and evolution, such as DNA replication and mRNA transcription, the binding of transcription factors (TFs) to TF binding sites (TFBS) plays a vital role. Precisely modeling the specificity of genes and searching for TFBS are helpful to explore the mechanism of cell expression. In recent years, computational and deep learning methods searching for TFBS have become an active field of research. However, existing methods generally cannot meet high performance and interpretability simultaneously. Here, we develop an accurate and interpretable attention-based hybrid approach, DeepARC, that combines a convolutional neural network (CNN) and recurrent neural network (RNN) to predict TFBS. DeepARC employs a positional embedding method to extract the hidden embedding from DNA sequences, including the positional information from OneHot encoding and the distributed embedding from DNA2Vec. DeepARC feeds the positional embedding of the DNA sequence into a CNN-BiLSTM-Attention-based framework to complete the task of finding the motif. Taking advantage of the attention mechanism, DeepARC can gain greater access to valuable information about the motif and bring interpretability to the work of searching for motifs through the attention weight graph. Moreover, DeepARC achieves promising performances with an average area under the receiver operating characteristic curve (AUC) score of 0.908 on five cell lines (A549, GM12878, Hep-G2, H1-hESC, and Hela) in the benchmark dataset. We also compare the positional embedding with OneHot and DNA2Vec and gain a competitive advantage.

| S-EPMC9204005 | biostudies-literature

Erroneous attribution of relevant transcription factor binding sites despite successful prediction of cis-regulatory modules.

Project description:BackgroundCis-regulatory modules are bound by transcription factors to regulate gene expression. Characterizing these DNA sequences is central to understanding gene regulatory networks and gaining insight into mechanisms of transcriptional regulation, but genome-scale regulatory module discovery remains a challenge. One popular approach is to scan the genome for clusters of transcription factor binding sites, especially those conserved in related species. When such approaches are successful, it is typically assumed that the activity of the modules is mediated by the identified binding sites and their cognate transcription factors. However, the validity of this assumption is often not assessed.ResultsWe successfully predicted five new cis-regulatory modules by combining binding site identification with sequence conservation and compared these to unsuccessful predictions from a related approach not utilizing sequence conservation. Despite greatly improved predictive success, the positive set had similar degrees of sequence and binding site conservation as the negative set. We explored the reasons for this by mutagenizing putative binding sites in three cis-regulatory modules. A large proportion of the tested sites had little or no demonstrable role in mediating regulatory element activity. Examination of loss-of-function mutants also showed that some transcription factors supposedly binding to the modules are not required for their function.ConclusionsOur results raise important questions about interpreting regulatory module predictions obtained by finding clusters of conserved binding sites. Attribution of function to these sites and their cognate transcription factors may be incorrect even when modules are successfully identified. Our study underscores the importance of empirical validation of computational results even when these results are in line with expectation.

| S-EPMC3235160 | biostudies-literature

A flexible integrative approach based on random forest improves prediction of transcription factor binding sites.

Project description:Transcription factor binding sites (TFBSs) are DNA sequences of 6-15 base pairs. Interaction of these TFBSs with transcription factors (TFs) is largely responsible for most spatiotemporal gene expression patterns. Here, we evaluate to what extent sequence-based prediction of TFBSs can be improved by taking into account the positional dependencies of nucleotides (NPDs) and the nucleotide sequence-dependent structure of DNA. We make use of the random forest algorithm to flexibly exploit both types of information. Results in this study show that both the structural method and the NPD method can be valuable for the prediction of TFBSs. Moreover, their predictive values seem to be complementary, even to the widely used position weight matrix (PWM) method. This led us to combine all three methods. Results obtained for five eukaryotic TFs with different DNA-binding domains show that our method improves classification accuracy for all five eukaryotic TFs compared with other approaches. Additionally, we contrast the results of seven smaller prokaryotic sets with high-quality data and show that with the use of high-quality data we can significantly improve prediction performance. Models developed in this study can be of great use for gaining insight into the mechanisms of TF binding.

| S-EPMC3413102 | biostudies-literature

Inclusion of neighboring base interdependencies substantially improves genome-wide prokaryotic transcription factor binding site prediction.

Project description:Prediction of transcription factor binding sites is an important challenge in genome analysis. The advent of next generation genome sequencing technologies makes the development of effective computational approaches particularly imperative. We have developed a novel training-based methodology intended for prokaryotic transcription factor binding site prediction. Our methodology extends existing models by taking into account base interdependencies between neighbouring positions using conditional probabilities and includes genomic background weighting. This has been tested against other existing and novel methodologies including position-specific weight matrices, first-order Hidden Markov Models and joint probability models. We have also tested the use of gapped and ungapped alignments and the inclusion or exclusion of background weighting. We show that our best method enhances binding site prediction for all of the 22 Escherichia coli transcription factors with at least 20 known binding sites, with many showing substantial improvements. We highlight the advantage of using block alignments of binding sites over gapped alignments to capture neighbouring position interdependencies. We also show that combining these methods with ChIP-on-chip data has the potential to further improve binding site prediction. Finally we have developed the ungapped likelihood under positional background platform: a user friendly website that gives access to the prediction method devised in this work.

| S-EPMC2896541 | biostudies-literature

Transfer learning with false negative control improves polygenic risk prediction.

Project description:Polygenic risk score (PRS) is a quantity that aggregates the effects of variants across the genome and estimates an individual's genetic predisposition for a given trait. PRS analysis typically contains two input data sets: base data for effect size estimation and target data for individual-level prediction. Given the availability of large-scale base data, it becomes more common that the ancestral background of base and target data do not perfectly match. In this paper, we treat the GWAS summary information obtained in the base data as knowledge learned from a pre-trained model, and adopt a transfer learning framework to effectively leverage the knowledge learned from the base data that may or may not have similar ancestral background as the target samples to build prediction models for target individuals. Our proposed transfer learning framework consists of two main steps: (1) conducting false negative control (FNC) marginal screening to extract useful knowledge from the base data; and (2) performing joint model training to integrate the knowledge extracted from base data with the target training data for accurate trans-data prediction. This new approach can significantly enhance the computational and statistical efficiency of joint-model training, alleviate over-fitting, and facilitate more accurate trans-data prediction when heterogeneity level between target and base data sets is small or high.

| S-EPMC10723713 | biostudies-literature

BERT-TFBS: a novel BERT-based model for predicting transcription factor binding sites by transfer learning.

Project description:Transcription factors (TFs) are proteins essential for regulating genetic transcriptions by binding to transcription factor binding sites (TFBSs) in DNA sequences. Accurate predictions of TFBSs can contribute to the design and construction of metabolic regulatory systems based on TFs. Although various deep-learning algorithms have been developed for predicting TFBSs, the prediction performance needs to be improved. This paper proposes a bidirectional encoder representations from transformers (BERT)-based model, called BERT-TFBS, to predict TFBSs solely based on DNA sequences. The model consists of a pre-trained BERT module (DNABERT-2), a convolutional neural network (CNN) module, a convolutional block attention module (CBAM) and an output module. The BERT-TFBS model utilizes the pre-trained DNABERT-2 module to acquire the complex long-term dependencies in DNA sequences through a transfer learning approach, and applies the CNN module and the CBAM to extract high-order local features. The proposed model is trained and tested based on 165 ENCODE ChIP-seq datasets. We conducted experiments with model variants, cross-cell-line validations and comparisons with other models. The experimental results demonstrate the effectiveness and generalization capability of BERT-TFBS in predicting TFBSs, and they show that the proposed model outperforms other deep-learning models. The source code for BERT-TFBS is available at https://github.com/ZX1998-12/BERT-TFBS.

| S-EPMC11066948 | biostudies-literature

Integrated assessment and prediction of transcription factor binding.

Project description:Systematic chromatin immunoprecipitation (chIP-chip) experiments have become a central technique for mapping transcriptional interactions in model organisms and humans. However, measurement of chromatin binding does not necessarily imply regulation, and binding may be difficult to detect if it is condition or cofactor dependent. To address these challenges, we present an approach for reliably assigning transcription factors (TFs) to target genes that integrates many lines of direct and indirect evidence into a single probabilistic model. Using this approach, we analyze publicly available chIP-chip binding profiles measured for yeast TFs in standard conditions, showing that our model interprets these data with significantly higher accuracy than previous methods. Pooling the high-confidence interactions reveals a large network containing 363 significant sets of factors (TF modules) that cooperate to regulate common target genes. In addition, the method predicts 980 novel binding interactions with high confidence that are likely to occur in so-far untested conditions. Indeed, using new chIP-chip experiments we show that predicted interactions for the factors Rpn4p and Pdr1p are observed only after treatment of cells with methyl-methanesulfonate, a DNA-damaging agent. We outline the first approach for consistently integrating all available evidences for TF-target interactions and we comprehensively identify the resulting TF module hierarchy. Prioritizing experimental conditions for each factor will be especially important as increasing numbers of chIP-chip assays are performed in complex organisms such as humans, for which "standard conditions" are ill defined.

| S-EPMC1479087 | biostudies-literature

Evaluating tools for transcription factor binding site prediction.

Project description:BackgroundBinding of transcription factors to transcription factor binding sites (TFBSs) is key to the mediation of transcriptional regulation. Information on experimentally validated functional TFBSs is limited and consequently there is a need for accurate prediction of TFBSs for gene annotation and in applications such as evaluating the effects of single nucleotide variations in causing disease. TFBSs are generally recognized by scanning a position weight matrix (PWM) against DNA using one of a number of available computer programs. Thus we set out to evaluate the best tools that can be used locally (and are therefore suitable for large-scale analyses) for creating PWMs from high-throughput ChIP-Seq data and for scanning them against DNA.ResultsWe evaluated a set of de novo motif discovery tools that could be downloaded and installed locally using ENCODE-ChIP-Seq data and showed that rGADEM was the best-performing tool. TFBS prediction tools used to scan PWMs against DNA fall into two classes - those that predict individual TFBSs and those that identify clusters. Our evaluation showed that FIMO and MCAST performed best respectively.ConclusionsSelection of the best-performing tools for generating PWMs from ChIP-Seq data and for scanning PWMs against DNA has the potential to improve prediction of precise transcription factor binding sites within regions identified by ChIP-Seq experiments for gene finding, understanding regulation and in evaluating the effects of single nucleotide variations in causing disease.

| S-EPMC6889335 | biostudies-literature

Universal prediction of cell cycle position using transfer learning

Project description:The cell cycle is a highly conserved, continuous process which controls faithful replication and division of cells. Single-cell technologies have enabled increasingly precise measurements of the cell cycle as both as a biological process of interest and as a possible confounding factor. Despite its importance and conservation, there is no universally applicable approach to infer position in the cell cycle with high-resolution from single-cell RNA-seq data. Here, we present tricycle, an R/Bioconductor package,to address this challenge by leveraging key features of the biology of the cell cycle, the mathematical properties of principal component analysis of periodic functions, and the ubiquitous applicability of transfer learning. We show that tricycle can predict any cell’s position in the cell cycle regardless of the cell type, species of origin, and even sequencing assay. The accuracy of tricycle compares favorably to gold-standard experimental assays which generally require specialized measurements in specifically constructed in vitro systems. Unlike gold-standard assays, tricycle is applicable to any single-cell RNA-seq dataset. Tricycle is highly scalable, universally accurate, and eminently pertinent for atlas-level data.

2021-04-07 | GSE171636 | GEO

Comparative genomics and disorder prediction identify biologically relevant SH3 protein interactions.

Project description:Protein interaction networks are an important part of the post-genomic effort to integrate a part-list view of the cell into system-level understanding. Using a set of 11 yeast genomes we show that combining comparative genomics and secondary structure information greatly increases consensus-based prediction of SH3 targets. Benchmarking of our method against positive and negative standards gave 83% accuracy with 26% coverage. The concept of an optimal divergence time for effective comparative genomics studies was analyzed, demonstrating that genomes of species that diverged very recently from Saccharomyces cerevisiae(S. mikatae, S. bayanus, and S. paradoxus), or a long time ago (Neurospora crassa and Schizosaccharomyces pombe), contain less information for accurate prediction of SH3 targets than species within the optimal divergence time proposed. We also show here that intrinsically disordered SH3 domain targets are more probable sites of interaction than equivalent sites within ordered regions. Our findings highlight several novel S. cerevisiae SH3 protein interactions, the value of selection of optimal divergence times in comparative genomics studies, and the importance of intrinsic disorder for protein interactions. Based on our results we propose novel roles for the S. cerevisiae proteins Abp1p in endocytosis and Hse1p in endosome protein sorting.

| S-EPMC1187863 | biostudies-literature

OmicsDI is part of the ELIXIR infrastructure

OmicsDI is an Elixir interoperability service. Learn more ›

Tweets

OmicsDI Databases

PRIDE
PeptideAtlas
MassIVE
JPOST Repository
Physiome Model Repository

EGA
EVA
ENA
LINCS
PAXDB
Cell Collective

MetaboLights
Metabolomics Workbench
MetabolomeExpress
GNPS
BioModels
FAIRDOMHub

ArrayExpress
dbGaP
ExpressionAtlas
GEO
NODE

Information

Databases
Help
API
Contact us
Code on GitHub
Terms of use
Submit Data