PRIME: Property-Informed Models#

Method Summary#

PRIME (Property-Informed Models of Evolution) is a framework of codon-level maximum likelihood methods that explicitly model amino acid exchangeability as a function of their physicochemical properties.

Standard codon substitution models represent selection using simple rate ratios ($dN/dS$ or $\omega$). While effective at identifying where and when selection occurs, these models are blind to the biophysical forces involved. For instance, standard models treat a mutation from Alanine to Valine (similar size/properties) identically to a mutation from Alanine to Arginine (large, charged).

PRIME solves this by parameterizing substitutions based on 5 core biophysical properties: 1. Molecular Volume 2. Hydropathy 3. Isoelectric Point (pI / Charge) 4. Alpha-Helix Propensity 5. Beta-Sheet Propensity

PRIME operates across three distinct levels of evolutionary resolution: * G-PRIME (Global): Models property constraints globally across the entire alignment and phylogeny. * E-PRIME (Episodic): Identifies lineage-specific changes in property constraints. * S-PRIME (Site-Level): Characterizes site-specific conservation or diversifying selection for specific properties.

What It Does#

• Resolves Biophysical Constraints: Identifies which specific physical properties (e.g. volume or charge) are being conserved or diversified at individual sites.
• Categorizes Site-Level Selection: S-PRIME groups sites into:
  • Property Conserved: Sites where amino acids are highly constrained to preserve a specific property.
  • Property Neutral: Sites evolving neutrally for that property.
  • Property Changing: Sites undergoing adaptive shifts in that property.
• Connects Evolution to Machine Learning: Demonstrates that PRIME selection weights align with the primary semantic axes of deep learning protein representations (like ESM-2) and agree with experimental deep mutational scanning (DMS) fitness landscapes.

How to Use It in HyPhy#

PRIME is implemented as a standard analysis template in HyPhy.

1. Prepare Input: You need a coding sequence alignment and an associated phylogenetic tree.
2. Execute PRIME: Run the analysis through the HyPhy command line: bash hyphy prime --alignment data.fas --tree tree.nwk
3. Choose Properties: The interactive prompt will allow you to select which physicochemical properties to model, or you can run with the default set of five properties.
4. Visualize Results: Output JSONs can be uploaded to HyPhy Vision to explore property constraints interactively via custom heatmaps.

Key Findings & Significance#

• Biophysical Realism: Modeling physical properties significantly improves statistical model fit compared to standard $dN/dS$ models.
• Biophysical Selection Hierarchy: Episodic analyses (E-PRIME) reveal that core packing and beta-sheet scaffolds are rigidly conserved, while alpha-helix propensity and surface electrostatics serve as the primary substrates for adaptive evolutionary tuning.
• High Sensitivity: The power to detect site-level property selection (S-PRIME) is governed by informational depth. It achieves $>90\%$ sensitivity in data-rich alignments.