Multiple graph regularized protein domain ranking

Background Protein domain ranking is a fundamental task in structural biology. Most protein domain ranking methods rely on the pairwise comparison of protein domains while neglecting the global manifold structure of the protein domain database. Recently, graph regularized ranking that exploits the global structure of the graph defined by the pairwise similarities has been proposed. However, the existing graph regularized ranking methods are very sensitive to the choice of the graph model and parameters, and this remains a difficult problem for most of the protein domain ranking methods. Results To tackle this problem, we have developed the Multiple Graph regularized Ranking algorithm, MultiG- Rank. Instead of using a single graph to regularize the ranking scores, MultiG-Rank approximates the intrinsic manifold of protein domain distribution by combining multiple initial graphs for the regularization. Graph weights are learned with ranking scores jointly and automatically, by alternately minimizing an ob- jective function in an iterative algorithm. Experimental results on a subset of the ASTRAL SCOP protein domain database demonstrate that MultiG-Rank achieves a better ranking performance than single graph regularized ranking methods and pairwise similarity based ranking methods. Conclusion The problem of graph model and parameter selection in graph regularized protein domain ranking can be solved effectively by combining multiple graphs. This aspect of generalization introduces a new frontier in applying multiple graphs to solving protein domain ranking applications.

💡 Research Summary

The paper tackles the problem of ranking protein domains—a fundamental task in structural biology where, given a query domain, one must retrieve the most similar domains from a large database. Traditional approaches rely on pairwise similarity measures such as TM‑score, DALI, or other structural alignment scores. While effective for individual comparisons, these methods ignore the global geometry of the entire domain collection, which can be viewed as a high‑dimensional manifold. Recent graph‑regularized ranking techniques address this limitation by constructing a single similarity graph over all domains and incorporating a Laplacian smoothness term that forces the ranking scores to vary smoothly along the graph. However, the performance of such methods is highly sensitive to how the graph is built: choice of distance metric, number of nearest neighbors (k), kernel bandwidth, and other hyper‑parameters dramatically affect the quality of the manifold approximation. Selecting optimal graph parameters typically requires extensive cross‑validation, which is computationally expensive and often impractical for large protein databases.

To overcome these difficulties, the authors propose MultiG‑Rank, a Multiple Graph regularized Ranking algorithm. Instead of committing to a single graph, MultiG‑Rank simultaneously employs a set of M initial graphs, each generated with a different combination of distance functions (e.g., Euclidean, cosine, Pearson correlation) and neighborhood definitions (k‑nearest‑neighbors, ε‑radius). All graphs share the same vertex set (the protein domains) but have distinct weighted adjacency matrices. The key idea is to learn a non‑negative weight vector w = (w₁,…,w_M) that linearly combines the Laplacians of these graphs, thereby approximating the true underlying manifold more faithfully than any single graph could.

The learning objective jointly optimizes the ranking scores f (a real‑valued vector assigning a relevance score to each domain) and the graph weights w:

\