[Paper Review] Computational identification of transcription factor binding sites by functional analysis of sets of genes sharing overrepresented upstream motifs

Davide Corà, Ferdinando Di Cunto+3 more|ArXiv.org|Oct 31, 2003

Genomics and Chromatin Dynamics20 references12 citations

TL;DR

This paper proposes a computational method to identify transcription factor binding sites by detecting overrepresented motifs in upstream gene regions and validating them through functional enrichment in Gene Ontology (GO) terms. By grouping genes sharing overrepresented motifs and testing for significant GO term enrichment, the approach identifies known and novel regulatory elements in *S. cerevisiae*, with a false discovery rate controlled via empirical randomization, enhancing reliability over expression-based methods alone.

ABSTRACT

BACKGROUND: Transcriptional regulation is a key mechanism in the functioning of the cell, and is mostly effected through transcription factors binding to specific recognition motifs located upstream of the coding region of the regulated gene. The computational identification of such motifs is made easier by the fact that they often appear several times in the upstream region of the regulated genes, so that the number of occurrences of relevant motifs is often significantly larger than expected by pure chance. RESULTS: To exploit this fact, we construct sets of genes characterized by the statistical overrepresentation of a certain motif in their upstream regions. Then we study the functional characterization of these sets by analyzing their annotation to Gene Ontology terms. For the sets showing a statistically significant specific functional characterization, we conjecture that the upstream motif characterizing the set is a binding site for a transcription factor involved in the regulation of the genes in the set. CONCLUSIONS: The method we propose is able to identify many known binding sites in S. cerevisiae and new candidate targets of regulation by known transcription factors. Its application to less well studied organisms is likely to be valuable in the exploration of their regulatory interaction network.

Motivation & Objective

• To improve the computational identification of transcription factor binding sites by leveraging functional annotation rather than relying solely on expression data.
• To address the limitations of microarray-based validation, such as experimental bias and low sensitivity for small gene sets.
• To develop a robust method for detecting biologically meaningful regulatory motifs using Gene Ontology (GO) term enrichment as a validation criterion.
• To control for false positives in motif-gene association by empirically estimating the false discovery rate using randomized gene sets.
• To extend the method’s utility to less-studied organisms by providing a functional validation framework independent of expression data.

Proposed method

• For all 5- to 8-nucleotide motifs, genes with significantly overrepresented motif occurrences in their upstream regions (relative to background frequencies) are grouped into motif-specific gene sets.
• Functional characterization of each motif-gene set is performed by testing for significant enrichment of Gene Ontology (GO) terms using a hypergeometric test.
• A false discovery rate (FDR) is estimated empirically by generating 3.5 million random gene sets of typical size (20 genes) and computing the distribution of best P-values across GO branches.
• The FDR is calculated as the ratio of expected false discoveries (from random sets) to observed significant associations, allowing cutoff selection for statistical confidence.
• The method uses a two-step validation: motif overrepresentation followed by functional coherence via GO enrichment, avoiding reliance on microarray data.
• The approach is applied to *S. cerevisiae*, with results compared to known transcription factor-gene interactions from the literature (e.g., Ref. [5]).

Experimental results

Research questions

• RQ1 Can functional enrichment in Gene Ontology terms serve as a reliable validation for computationally identified transcription factor binding motifs?
• RQ2 How can false positive motif-gene associations be controlled when testing thousands of motifs and GO terms simultaneously?
• RQ3 Does GO-based validation detect regulatory motifs that are missed by expression-based methods due to experimental limitations?
• RQ4 To what extent can this method identify known and novel regulatory elements in a well-annotated organism like *S. cerevisiae*?
• RQ5 Can this approach be generalized to less-studied organisms where expression data may be limited or unavailable?

Key findings

• The method successfully identifies many known transcription factor binding sites in *S. cerevisiae* through significant functional enrichment of GO terms in motif-associated gene sets.
• The approach detects novel candidate regulatory targets for known transcription factors by linking overrepresented motifs to coherent biological functions.
• By using empirical randomization, the method controls the false discovery rate at 0.01, ensuring high confidence in the identified motif-gene associations.
• The GO-based validation complements and extends expression-based methods, particularly for smaller or functionally coherent gene sets that may not yield strong microarray signals.
• The method demonstrates robustness across different set sizes, with simulations showing consistent FDR estimates regardless of the chosen gene set size.
• A systematic comparison with experimentally determined transcription factor targets (from Ref. [5]) shows that motif-gene sets with significant GO enrichment have highly significant overlaps with known TF-regulated genes (P < 10⁻⁵ for many cases).