High ColabFold Confidence Does Not Guarantee Catalytic-Site Accuracy in Bacillus subtilis PdxT: ColabFold Confidence vs Catalytic-Site Accuracy in Bacillus subtilis PdxT

Mateen Ur Rehman; Sheheryar Ahmad Khan; Bisma Azam; Jannat Bibi; Amna Bibi; Muhammad Abu Baker; Nida Shabbir

doi:10.54393/fbt.v6i1.228

Authors

Mateen Ur Rehman Institute of Molecular Biology and Biotechnology, The University of Lahore, Lahore, Pakistan
Sheheryar Ahmad Khan Institute of Molecular Biology and Biotechnology, The University of Lahore, Lahore, Pakistan
Bisma Azam Institute of Molecular Biology and Biotechnology, The University of Lahore, Lahore, Pakistan
Jannat Bibi Institute of Molecular Biology and Biotechnology, The University of Lahore, Lahore, Pakistan
Amna Bibi Institute of Molecular Biology and Biotechnology, The University of Lahore, Lahore, Pakistan
Muhammad Abu Baker Institute of Molecular Biology and Biotechnology, The University of Lahore, Lahore, Pakistan
Nida Shabbir Institute of Molecular Biology and Biotechnology, The University of Lahore, Lahore, Pakistan

DOI:

https://doi.org/10.54393/fbt.v6i1.228

Keywords:

PdxT, Pyridoxal Phosphate, Vitamin B6 Biosynthesis, Protein Structure Prediction, Colabfold, Structural Validation, Antimicrobial Drug Target

Abstract

AI-based protein structure predictors such as AlphaFold2 and ColabFold routinely generate models with high confidence scores for backbone geometry. However, whether these global metrics reliably capture catalytically competent active-site configurations in enzymes remains unclear. Objective: To evaluate whether a high-confidence ColabFold model of the glutaminase subunit PdxT from Bacillus subtilis accurately reproduces the geometry of its catalytic cysteine–histidine–glutamate triad. Methods: The amino acid sequence of B. subtilis PdxT (UniProt P37528) was submitted to ColabFold v1.5. Five models were generated with default settings, AMBER relaxation, and MMseqs2-based multiple sequence alignment. The top-ranked model was selected based on predicted Local Distance Difference Test (pLDDT) and predicted TM-score (pTM). Inter-residue distances between Cys118, His168, and Glu51 were measured from the predicted structure and compared with distance ranges (2.5–5.0 Å) reported for experimentally solved Class I glutaminase structures. Results: The top ColabFold model displayed high global confidence (mean pLDDT 96.4; pTM 0.929). The measured inter-residue distances were 10.36 Å (Cys118–His168) and 18.0 Å (His168–Glu51), exceeding the 2.5–5.0 Å range typically required for catalytic function. No experimental validation or additional computational analyses were performed. Conclusions: In this PdxT model, high global confidence metrics did not correspond to catalytically realistic active-site geometry. These findings suggest that AI-generated protein models intended for functional interpretation may require secondary validation focused on active-site architecture.

References

Yuan Y and Chen L. Transporters in Vitamin Uptake and Cellular Metabolism: Impacts on Health and Disease. Life Metabolism. 2025 Jun; 4(3): 08. doi: 10.1093/lifemeta/loaf008.

Denise R, Babor J, Gerlt JA, de Crécy-Lagard V. Pyridoxal 5’-Phosphate Synthesis and Salvage in Bacteria and Archaea: Predicting Pathway Variant Distributions and Holes. Microbial Genomics. 2023 Feb; 9(2): 926. doi: 10.1099/mgen.0.000926.

Mapelli-Brahm P, Gómez-Villegas P, Gonda ML, León-Vaz A, León R, Mildenberger J, Rebours C, Saravia V, Vero S, Vila E, Meléndez-Martínez AJ. Microalgae, Seaweeds and Aquatic Bacteria, Archaea, and Yeasts: Sources of Carotenoids with Potential Antioxidant and Anti-Inflammatory Health-Promoting Actions in the Sustainability Era. Marine Drugs. 2023 Jun; 21(6): 340. doi: 10.3390/md21060340.

Seshadri K, Abad AN, Nagasawa KK, Yost KM, Johnson CW, Dror MJ et al. Synthetic Biology in Natural Product Biosynthesis. Chemical reviews. 2025 Mar; 125(7): 3814-3931. doi: 10.1021/acs.chemrev.4c00567.

Ito T, Ogawa H, Hemmi H, Downs DM, Yoshimura T. Mechanism of Pyridoxine 5′-Phosphate Accumulation in Pyridoxal 5′-Phosphate-Binding Protein Deficiency. Journal of Bacteriology. 2022 Mar; 204(3): 21. doi: 10.1128/jb.00521-21.

Richts B and Commichau FM. Underground Metabolism Facilitates the Evolution of Novel Pathways for Vitamin B6 Biosynthesis. Applied Microbiology and Biotechnology. 2021 Mar; 105(6): 2297-2305. doi: 10.1007/s00253-021-11199-w.

Pina AF, Sousa SF, Cerqueira NM. The Catalytic Mechanism of Pdx2 Glutaminase Driven by a Cys‐His‐Glu Triad: A Computational Study. Chem-Bio-Chem. 2022 May; 23(9): 202100555. doi: 10.1002/cbic.202100555.

Dutta D, Mishra S. The Structural and Energetic Aspects of Substrate Binding and the Mechanism of Action of the DapE-Encoded N-Succinyl-L, L-Diaminopimelic Acid Desuccinylase (DapE) Investigated Using a Hybrid QM/MM Method. Physical Chemistry Chemical Physics. 2014; 16(47): 26348-26358. doi: 10.1039/C4CP03986F.

Novikova IV, Zhou M, Du C, Parra M, Kim DN, VanAernum ZL et al. Tunable Heteroassembly of a Plant Pseudoenzyme–Enzyme Complex. ACS Chemical Biology. 2021 Sep; 16(11): 2315-2325. doi: 10.1021/acschembio.1c00475.

Bossé JT, Bujold AR, Li L. Actinobacillus. Pathogenesis of Bacterial Infections in Animals. 2022 Dec: 262-289. doi: 10.1002/9781119754862.ch12.

Koshak YF, Benedykt VV, Prodan AM, Dzhyvak VG, Nikitina IM. Causes of Superinfections: Deadly Bacteria of Tuberculosis. Infection. 2022; 1(2). 1-9.

Richts B, Lentes S, Poehlein A, Daniel R, Commichau FM. A Bacillus subtilis Δ pdxT Mutant Suppresses Vitamin B6 Limitation by Acquiring Mutations Enhancing the Pdxs Gene Dosage and Ammonium Assimilation. Environmental Microbiology Reports. 2021 Apr; 13(2): 218-233. doi: 10.1111/1758-2229.12936.

Ali NF, Khan S, Zahid S. A Critical Address to Advancements and Challenges in Computational Strategies for Structural Prediction of Protein in the Recent Past. Computational Biology and Chemistry. 2025 Aug; 117: 108430. doi: 10.1016/j.compbiolchem.2025.108430.

Zhang J and Qian J. Advances in Computational Intelligence-Based Methods of Structure and Function Prediction of Proteins. Biomolecules. 2024 Aug; 14(9): 1083. doi: 10.3390/biom14091083.

Chen L, Li Q, Nasif KF, Xie Y, Deng B, Niu S et al. AI-Driven Deep Learning Techniques in Protein Structure Prediction. International Journal of Molecular Sciences. 2024 Aug; 25(15): 8426. doi: 10.3390/ijms25158426.

Schauperl M and Denny RA. AI-Based Protein Structure Prediction in Drug Discovery: Impacts and Challenges. Journal of Chemical Information and Modeling. 2022 Jun; 62(13): 3142-3156. doi: 10.1021/acs.jcim.2c00026.

Ozden B, Kryshtafovych A, Karaca E. The Impact of AI‐Based Modeling on The Accuracy of Protein Assembly Prediction: Insights from CASP15. Proteins: Structure, Function, and Bioinformatics. 2023 Dec; 91(12): 1636-1657. doi: 10.1002/prot.26598.

Modi M, Thambiraja M, Cherukat A, Yennamalli RM, Priyadarshini R. Structure Predictions and Functional Insights into Amidase_3 Domain Containing N-Acetylmuramyl-L-Alanine Amidases from Deinococcus Indicus DR1. BioMed Central Microbiology. 2024 Mar; 24(1): 101. doi: 10.1186/s12866-024-03225-4.

Krivitskaya AV, Khrenova MG, Nemukhin AV. Two Sides of Quantum-Based Modeling of Enzyme-Catalyzed Reactions: Mechanistic and Electronic Structure Aspects of the Hydrolysis by Glutamate Carboxypeptidase. Molecules. 2021 Oct; 26(20): 6280. doi: 10.3390/molecules26206280

Qiu W, Yang P, Ye J, Zhou J, Liu S. Unveiling Highly Active and Stable L-Glutaminase Through Ancestral Sequence Reconstruction and Turnover Number Prediction. Journal of Agricultural and Food Chemistry. 2025 Feb; 73(9): 5353-5362. doi: 10.1021/acs.jafc.4c11502.