Morpheus: A fragment-based algorithm to predict metamorphic behaviour in proteins across proteomes

Vijay Subramanian, Rajeswari Appadurai, H.n. Venkatesh, Ashok Sekhar, Anand SrivastavabioRxiv 2025 · DOI

TL;DR

Researchers have developed "Morpheus," a novel fragment-based algorithm that predicts whether a protein can adopt multiple stable structures (metamorphic behavior) solely from its amino acid sequence. This tool was applied to over 600,000 proteins across 57 proteomes, identifying approximately 10% as potentially metamorphic, and offers a web server for broader use.

What Problem Does This Paper Address?

The classical understanding of protein folding is that a single amino acid sequence dictates a single, stable three-dimensional structure. However, a class of proteins known as "metamorphic proteins" defy this dogma by being able to adopt two or more distinct, stable conformations, often with different functions. These "fold-switching" proteins are thought to be widespread, but identifying them computationally has been challenging. Existing methods often rely on complex structural data or are limited in their ability to screen entire proteomes. This paper aims to bridge this gap by providing an accurate, sequence-based method to predict metamorphic behavior across a large scale.

Methods

Morpheus employs a fragment-based approach. It breaks down a query protein sequence into overlapping 7-residue fragments. Each fragment is then searched against a comprehensive database of known protein structures (from PDB and AlphaFold) to find similar fragments. The key innovation lies in analyzing the diversity of secondary structures adopted by these matching fragments.

Four metrics are calculated for each fragment:

  • Diversity Index: Quantifies the variation in secondary structure predictions.
  • Substitution Score: Measures how well a fragment's secondary structure aligns with a reference structure.
  • Information Entropy: Assesses the uncertainty in secondary structure prediction.
  • Uncertainty Score: Estimates the confidence in the calculated diversity metrics based on the number of matching fragments.

These four metrics, along with a rolling average to capture contiguous regions of high diversity, are used as features for a Support Vector Machine (SVM) classifier. The SVM is trained to distinguish between metamorphic and monomorphic proteins. The authors also optimized fragment size, the residue within the fragment to focus on (the 3rd residue), and the substitution matrix parameters.

Datasets

  • Training Data: A curated dataset of 189 known fold-switching proteins and 198 monomorphic proteins, filtered to remove obsolete or corrected structures. This dataset was expanded to include designed metamorphic proteins.
  • Fragment Database: Approximately 200,000 redundant proteins from the Protein Data Bank (PDB) and around 1.2 million protein structure predictions from AlphaFold2.

Key Insights and Ideas

  • Sequence alone can hint at metamorphic potential: The core idea is that the variability in secondary structure adoption by short sequence fragments, when compared against a large structural database, is a strong indicator of a protein's ability to switch folds.
  • Diversity metrics capture fold-switching regions: Regions within a protein exhibiting high entropy scores (a measure of uncertainty in secondary structure prediction) often correspond to experimentally identified regions of fold-switching.
  • Fragment-based approach scales well: By using a Trie data structure for fragment searching, Morpheus can efficiently process large proteomes.
  • Metamorphic proteins are more prevalent than previously thought: The study's application to 57 proteomes suggests that about 10% of proteins might possess metamorphic capabilities, a higher estimate than some previous figures.
  • Intrinsically Disordered Proteins (IDPs) can exhibit high diversity: The algorithm flags IDPs as potentially metamorphic due to their inherent conformational flexibility, highlighting the continuum between ordered, metamorphic, and disordered states.

Key Results

  • High Accuracy: Morpheus achieves an average cross-validation accuracy of 84.7% with a Matthew's Correlation Coefficient (MCC) of 0.698.
  • Proteome-wide Screening: The algorithm was successfully applied to 57 proteomes, totaling 601,218 proteins, identifying approximately 10% (around 60,000 proteins) as potentially metamorphic.
  • Web Server Available: A web server is provided for users to test their own protein sequences.
  • Identification of Candidate Proteins: The study identified several candidate metamorphic proteins involved in crucial biological processes, suggesting avenues for further experimental investigation.

Limitations and Future Work

  • Sensitivity to Point Mutations: Morpheus may struggle to predict metamorphic behavior when sequences differ only by point mutations, as demonstrated by the example of E48A in RfaH.
  • Tertiary Structure Rearrangements: Proteins that switch folds primarily through tertiary contact or hydrogen bonding rearrangements, without significant secondary structure changes, might be missed.
  • Disorder Prediction: While the algorithm includes IDPs, further refinement using dedicated disorder prediction tools could help filter the metamorphic set more precisely.
  • Experimental Validation: The authors emphasize that their predictions are computational and require experimental validation.
  • Future Work: The authors suggest that their method will be useful for de novo design and engineering of metamorphic proteins. They also highlight the potential for further research into specific candidate proteins identified in their study, particularly those involved in circadian rhythms or with no solved structures.

Summary updated

Abstract

Abstract Functionally important “fold-switching” proteins, which do not obey the classical folding dogma, are now thought to be widespread. Algorithms that can accurately annotate fold-switching proteins from sequence information can help uncover the true extent of the “metamorphome”. Here, we present Morpheus, a fragment-based classification approach, that works by analysing the diversity of structures within a query protein sequence. Morpheus exhaustively curates and uses fragment structural data from the protein data bank as well as the AlphaFold Protein Structure Database. We employed our algorithm on 57 different proteomes consisting of a total of 601,218 proteins and identified about 10% of these proteins with the ability to fold switch. Additionally, we provide a web server for Morpheus to test for metamorphic propensities for user-defined sequences ( http://mbu.iisc.ac.in/∼anand/morpheus ). Besides screening for metamorphic behaviour in proteomes, our work will be useful in de novo design and engineering of such proteins through further experimentation.
BibTeX
@misc{subramanian2025morpheus,
  author = {Vijay Subramanian and Rajeswari Appadurai and H.N. Venkatesh and Ashok Sekhar and Anand Srivastava},
  title = {{Morpheus: A fragment-based algorithm to predict metamorphic behaviour in proteins across proteomes}},
  year = {2025},
  doi = {10.1101/2025.02.13.637956},
  howpublished = {bioRxiv},
}

Loading related content...

Figure