Issue with NaN Values in PDB Output When Running AF_unmasked with Limited MSA

[Ranim-Benkhlifa]

Hello,

Thank you for your work on this tool, it's greatly appreciated!

I'm currently using the AF_unmasked script for a research project to assess the impact of minimal evolutionary information in the MSA input, specifically focusing on cases where the MSA consists solely of the target protein sequence without any homologous sequences.
While the script executes without any runtime errors, the resulting PDB files contain NaN values in atomic coordinates (X, Y, Z) and occupancy fields, rendering the structures unusable for visualization or downstream analysis. This issue consistently occurs when the MSA lacks homologous sequences.
Could you please provide guidance on whether this behavior is an expected limitation when the MSA is sparse? Additionally, are there any recommended adjustments to still generate structurally valid predictions, even with low-confidence scores?

Thank you for your time and for sharing such a valuable resource.

Best regards,
Ranim.

[clami66]

Dear Ranim,

This sounds strange, and potentially hinting at an issue with your installation. Do you have an example I could run on my system to reproduce the issue?