Can GATE be used for monomer model selection

[YujieMo]

Hi,

Thank you for the interesting work. Seems like the GATE is mainly designed for assessing the multimer, can it be used for monomer and how about the performance on monomer?

[JianLiu1994]

Hi Yujie,

Yes, GATE can be used to assess monomer structures. We have updated the instructions and code to support evaluating the quality of a pool of protein tertiary structure models.

GATE was trained on our in-house CASP15 monomer structures predicted by AlphaFold2, and it outperforms all the quality scores used as node features in the pairwise similarity graph (including AlphaFold pLDDT, APOLLO, EnQA, GCPNet-EMA, and DeepRank3) on our CASP15 in-house test set.

We also included results on the CASP16 in-house dataset in our recent paper: Boosting AlphaFold Protein Tertiary Structure Prediction through Extensive Model Sampling and Ranking (https://www.biorxiv.org/content/10.1101/2025.06.06.658338v1.full).

Please let us know if you encounter any issues or run into problems while using GATE.

[YujieMo]

Hi,

Thank you for your timely response. Actually, I have many models predicted by different methods, such as alphafold2, boltz-2, and so on. Can GATE be used to evaluate the models from different predicted methods? Or, are there any other methods can be used to do such things? That would be very helphful for our research.

Thanks.

[JianLiu1994]

Hi Yujie,

Yes, GATE can be applied to models generated by different methods, as long as the b-factor column contains values corresponding to pLDDT. GATE extracts the pLDDT scores from the b-factor column and uses them as features. Another note is that running GATE on a large pool of models can be time-consuming, since the PSS feature takes considerable time to generate.

[YujieMo]

Hi Jian,

Thank you for your response and update on the code for monomer model selection. But when I download the code and run it, I found that some functions in the code are still missed. For example, the sample_models_by_kmeans_monomer() and features_monomer_dict(). Could you provide these functions so I can run them on the singleton model?

Thanks.

[YujieMo]

Hi Jian,

Another issue is the implementation of DeepRank3, when I conduct the perl setup_database.pl, some links of the database cannot be downloaded. So, if we cannot successfully implement the DeepRank3, can GATE still works?

Thanks.

[JianLiu1994]

Hi Yujie,

I have updated the code for running the monomer model selection. For the DeepRank3, it's a very old implementation, so I am working on restoring the tools/databases for you to download. However, some of the tools are pretty old, so I cannot guarantee that they can be installed on your computer.

I will let you know once I have finished restoring the tools/databases required to run DeepRank3.

Thanks

[JianLiu1994]

Hi Yujie,

I have updated the installation codes for DeepRank3 and restored the necessary tools and databases. Since this is an older implementation, some tools may not work on all systems.

For reference, the tools and databases were tested and restored on the following system:

CPU: Intel(R) Xeon(R) Gold 6254 @ 3.10GHz (48 cores)
OS: AlmaLinux 8.10 (Cerulean Leopard)
Kernel: Linux 5.14.0-427.13.1.el8_10.x86_64
GCC: 8.5.0 (Red Hat 8.5.0-28)
glibc: 2.28

Please let me know if you encounter any issues during installation or while running DeepRank3.

Thanks,
Jian