Real differences or modeling noise? Two tilapia virus variants, checked with MolProbity
Michigan State researchers used MolProbity on ProteinIQ to confirm that predicted PB2 models of two Egyptian tilapia virus variants were sound before comparing them.
Matic Broz, PhD2 min

- Authors
- El Asely A.M., Khalifa M.S., Xu W., Shaheen A.A., et al.
- Published in
- Microorganisms · 2026-02 · Michigan State University
- ProteinIQ in this study
- Predicted PB2 protein structures were validated with ProteinIQ's MolProbity tool.
- Tools used
- Molprobity
Two viruses from farmed Egyptian tilapia looked different from the tilapia lake virus reference. The researchers had a prediction for how, but first they had to rule out a duller explanation: that the differences came from the way the models were built.
Researchers at Michigan State University and Benha University sequenced two variants of a divergent amnoonvirus, AmnoonvirusEGY1F and AmnoonvirusEGY1H, from tilapia in Egypt. Tilapia lake virus (TiLV) has threatened tilapia aquaculture since 2006, so knowing how far the variants drift matters for surveillance. The team assembled segment 2, which encodes the PB2 subunit of the viral RNA polymerase, an enzyme central to replication and immune evasion.
Checking the models before trusting the differences
There is no experimental structure for these proteins. The team built 3D models of all three PB2 proteins with SWISS-MODEL, then faced a familiar trap. A model can look fine in a viewer and still contain geometric errors, and structural differences between strains could just as easily reflect modeling artifacts as real biology.
They ran all three models through MolProbity. It combines the clash score, Ramachandran statistics and rotamer outliers into one MolProbity score that can be compared with experimental crystal structures. The authors describe it this way in their methods:
This valuable tool collates multifaceted computational analyses to check for the overall geometric correctness of the predicted 3D protein structure and combine them into one value: the MolProbity score.
Three scores, then a map of differences
All three models scored between 1 and 2, the range of high-resolution crystal structures: 1.11 for AmnoonvirusEGY1F, 1.12 for the TiLV index strain and 1.24 for AmnoonvirusEGY1H. With the models cleared, the team used MolProbity's per-residue bond angle output to find where the strains diverge.
Deviations at six positions (C55 to C57, C102 to C103, C141 to C142, C317 to C318, C324 to C325 and C331 to C332) appeared in the TiLV index strain but in neither Egyptian variant. Deviations at C144 and C147 appeared only in the Egyptian strains.
What separates the variants
Sequence analysis found 28 amino acid substitutions between the two Egyptian contigs. Two are unique to EGY1H, A230E and A231T, and SIFT and ConSurf predicted both to matter functionally because they sit in a highly conserved region. FoldX energies pointed the same way, at 292.3 kcal/mol for TiLV, 282.89 for EGY1F and 277.86 for EGY1H.
The authors conclude that the Egyptian variants are genetically and structurally distinct from known TiLV strains. That bears directly on how diagnostic assays are designed and on what animal health authorities decide to report.


