MOTIPS: Automated Motif Analysis for Predicting Targets of Modular Protein Domains

Hugo Y. K. Lam, Philip M. Kim, Janine Mok, Raffi Tonikian, Sachdev S. Sidhu, Benjamin E. Turk, Michael Snyder, Mark B. Gerstein

Introduction

Identifying protein-protein interactions (PPIs) is fundamental to mapping cellular signaling pathways. Many of these interactions are mediated by modular domains that recognize short, linear motifs (SLiMs), typically 5–10 amino acids long. However, traditional methods that scan proteomes for sequence matches often suffer from high false-positive rates. A sequence might match a known binding pattern but be biologically inaccessible—buried within a protein's hydrophobic core or locked in a rigid structure.

MOTIPS (Motif-based Interaction Prediction System) addresses this by looking beyond the primary sequence. By integrating structural context—such as surface exposure and intrinsic disorder—with evolutionary conservation data, MOTIPS significantly improves the accuracy of target prediction.

The MOTIPS Pipeline

The authors developed MOTIPS as an automated search and integration algorithm designed to formalize the "expert eye" of a biologist. The pipeline operates in four distinct stages:

  1. Normalization: Raw experimental data from sources like phage display or synthetic peptide libraries are converted into standardized Position Specific Scoring Matrices (PSSMs).
  2. Motif Scanning: A sliding-window approach scans target proteomes for potential matches.
  3. Feature Integration: For every potential hit, MOTIPS calculates three "genomic" features: Conservation (via AL2CO), Surface Propensity (via SABLE), and Disorder (via DISOPRED).
  4. Classification: A Naive Bayes classifier integrates these features and the number of hits per protein to calculate the likelihood of a true biological interaction.
Experimental data generation via phage display for motif identification, showing the selection and sequencing of binding clones.

Datasets and Benchmarking

The system was trained using experimentally determined interaction pairs for the SH3 domain (Sho1 protein) and the Serine/Threonine kinase domain (Prk1). To ensure robustness, the researchers utilized "Gold Standard Negatives"—protein pairs known not to interact due to differing cellular localizations. The pipeline was pre-loaded with data for 20 proteomes, including yeast, worms, and mammals.

Example of a peptide mixture array used for S/T kinase screening, where specific amino acids are fixed at one position while others remain degenerate.

Key Insights: Why Context Matters

The core innovation of MOTIPS is the realization that a motif's physical environment is as critical as its sequence. Principal Component Analysis (PCA) revealed that structural features (like disorder) and evolutionary features (conservation) provide independent, non-redundant information. This independence justifies the use of a Naive Bayes model and explains why integrating these disparate data types leads to a dramatic performance boost.

Performance Results

MOTIPS demonstrated substantial improvements over existing tools like the Eukaryotic Linear Motif (ELM) database and Scansite:

  • Accuracy: MOTIPS achieved a 49% increase in accuracy over ELM for SH3 targets and was 90% more accurate than Scansite's profile-scoring method.
  • Predictive Power: For Prk1 kinase targets, the Area Under the Curve (AUC) rose from 0.72 (sequence only) to 0.79 with integrated features.
  • Scalability: The algorithm remains computationally efficient, capable of scanning large mammalian proteomes by focusing on a prioritized list of high-scoring hits.

Limitations and Future Directions

While a significant advancement, MOTIPS' performance is inherently tied to the quality of its training data. Future iterations could benefit from expanding the feature set to include protein-protein interaction network topology or more granular cellular localization data. Additionally, as the relationships between structural features become better understood, more complex machine learning architectures like Random Forests or Neural Networks may eventually supersede the Naive Bayes approach.

Source: Lam, H. Y. K., et al. (2010). MOTIPS: Automated Motif Analysis for Predicting Targets of Modular Protein Domains. BMC Bioinformatics, 11:243.

Summary updated

Abstract

Background Many protein interactions, especially those involved in signaling, involve short linear motifs consisting of 5-10 amino acid residues that interact with modular protein domains such as the SH3 binding domains and the kinase catalytic domains. One straightforward way of identifying these interactions is by scanning for matches to the motif against all the sequences in a target proteome. However, predicting domain targets by motif sequence alone without considering other genomic and structural information has been shown to be lacking in accuracy. Results We developed an efficient search algorithm to scan the target proteome for potential domain targets and to increase the accuracy of each hit by integrating a variety of pre-computed features, such as conservation, surface propensity, and disorder. The integration is performed using naïve Bayes and a training set of validated experiments. Conclusions By integrating a variety of biologically relevant features to predict domain targets, we demonstrated a notably improved prediction of modular protein domain targets. Combined with emerging high-resolution data of domain specificities, we believe that our approach can assist in the reconstruction of many signaling pathways.
BibTeX
@article{lam2010motips,
  author = {Hugo Y. K. Lam and Philip M. Kim and Janine Mok and Raffi Tonikian and Sachdev S. Sidhu and Benjamin E. Turk and Michael Snyder and Mark B. Gerstein},
  title = {{MOTIPS: Automated Motif Analysis for Predicting Targets of Modular Protein Domains}},
  journal = {BMC Bioinformatics},
  year = {2010},
  doi = {10.1186/1471-2105-11-243},
}

Loading related content...

Figure