Skip to content
DraconiticResearch
Search Sources API
194,226 studies31 sourcesUpdated 22 min ago
Peer-reviewedCC BY 4.0Molecules

TMK-VS-Pred: Impact of Decoy Choice on Target-Specific Scoring for Structure-Based Virtual Screening of Thymidylate Kinase, a Potential Anti-Mycobacterium Tuberculosis Target

  • Muhammad Junaid1,2
  • Muhammad Zeeshan3
  • Abbas Khan4
  • Wenjin Li1
4 affiliations
  1. 1Institute for Advanced Study, Shenzhen University, Shenzhen 518060, China
  2. 2School of Life Science and Biotechnology, Shanghai Jiao Tong University, 800 Dongchuan Rd., Minhang District, Shanghai 200240, China
  3. 3Department of Bioinformatics and Biotechnology, Islamic International University Islamabad, Islamabad 44000, Pakistan
  4. 4Department of Biomedical Sciences, School of Medical and Life Sciences, Sunway University, Sunway City 47500, Malaysia
Open at sourceDOICite

Abstract

Artificial intelligence (AI) has revolutionized drug discovery by enhancing the efficiency and accuracy of structure-based virtual screening (SBVS) and binding affinity prediction. SBVS relies on scoring functions (SFs) to identify potential drug candidates, but conventional SFs often lack target specificity. AI-driven SFs overcome these limitations through data-driven approaches, improving predictive performance. Target-specific machine learning SFs (MLSFs) further enhance binding affinity predictions by capturing intricate molecular interactions, leading to more reliable virtual screening outcomes. This study focuses on thymidylate kinase (TMK) of Mycobacterium tuberculosis, a pivotal enzyme in nucleotide synthesis and DNA replication, making it a promising target for the development of new anti-TB therapies. Among the trained MLSFs, the ANN-PLEC classifier emerged as the top-performing model, achieving mean PR-AUC scores of 0.90. The corresponding F1 scores were 0.88. Our findings reveal that training and testing models on the same type of decoys introduces decoy bias, leading to overestimated SBVS performance. Importantly, ANN-PLEC demonstrated superior classification performance, achieving an NEF1% of 0.96. Additionally, XGB-GRID excelled in binding affinity prediction, with Pearson and Spearman correlation coefficients of 0.87 and 0.73, respectively, showcasing its capability to rank molecules accurately by binding affinity.

Related

  1. Crystal structure of the bifunctional enzyme phosphoribosyl isomerase A from Mycobacterium tuberculosis with bound light-controllable inhibitor
    Chitra Rajendran, Reinhard Sterner
    Acta Crystallographica Section F: Structural Biology CommunicationsPeer-reviewedCC BY 4.0Full textMaterials & polymersPMC
  2. Structure-based virtual screening identifies naturally derived histidine kinase VbrK inhibitors as potential countermeasures against β-lactam antimicrobial resistance
    Chao Yan, Xiaoyi Li, Jiayin Wang +3
    RSC Medicinal ChemistryPeer-reviewedPublisher termsFull textGeneral chemistryPMC
  3. Structural basis of interaction between Pol1 and DnaN from Mycobacterium tuberculosis
    D. Sharma, D. T. Nair
    bioRxivPreprintAll rights reservedFull textMaterials & polymersbioRxiv
  4. Identification and Experimental Validation of Repurposable Drugs Targeting the Essential ClpP1 Protease of Mycobacterium tuberculosis
    Smriti Bhardwaj, Kuldeep K. Roy
    ChemRxivPreprintLicence not statedGeneral chemistryChemRxiv
  5. Cryo-EM structures of Mycobacterium tuberculosis pyruvate carboxylase reveal acetyl-CoA-mediated tetramer stabilization and carrier-domain coordination
    Aekagra Singh, Sandeep Singh, Uddipan Das
    Communications BiologyPeer-reviewedCC BY-NC-ND 4.0Full textMaterials & polymersCrossref

Cite

BibTeXRIS
BibTeX
@article{junaid2026tmkvspred,
  title = {{TMK-VS-Pred: Impact of Decoy Choice on Target-Specific Scoring for Structure-Based Virtual Screening of Thymidylate Kinase, a Potential Anti-Mycobacterium Tuberculosis Target}},
  author = {Muhammad Junaid and Muhammad Zeeshan and Abbas Khan and Wenjin Li},
  journal = {Molecules},
  year = {2026},
  month = oct,
  doi = {10.3390/molecules31213834},
  url = {https://www.mdpi.com/1420-3049/31/21/3834},
  note = {CC BY 4.0},
}
RIS
TY  - JOUR
TI  - TMK-VS-Pred: Impact of Decoy Choice on Target-Specific Scoring for Structure-Based Virtual Screening of Thymidylate Kinase, a Potential Anti-Mycobacterium Tuberculosis Target
AU  - Muhammad Junaid
AU  - Muhammad Zeeshan
AU  - Abbas Khan
AU  - Wenjin Li
PY  - 2026
DA  - 2026/10/23
JO  - Molecules
DO  - 10.3390/molecules31213834
UR  - https://www.mdpi.com/1420-3049/31/21/3834
AB  - Artificial intelligence (AI) has revolutionized drug discovery by enhancing the efficiency and accuracy of structure-based virtual screening (SBVS) and binding affinity prediction. SBVS relies on scoring functions (SFs) to identify potential drug candidates, but conventional SFs often lack target specificity. AI-driven SFs overcome these limitations through data-driven approaches, improving predictive performance. Target-specific machine learning SFs (MLSFs) further enhance binding affinity predictions by capturing intricate molecular interactions, leading to more reliable virtual screening outcomes. This study focuses on thymidylate kinase (TMK) of Mycobacterium tuberculosis, a pivotal enzyme in nucleotide synthesis and DNA replication, making it a promising target for the development of new anti-TB therapies. Among the trained MLSFs, the ANN-PLEC classifier emerged as the top-performing model, achieving mean PR-AUC scores of 0.90. The corresponding F1 scores were 0.88. Our findings reveal that training and testing models on the same type of decoys introduces decoy bias, leading to overestimated SBVS performance. Importantly, ANN-PLEC demonstrated superior classification performance, achieving an NEF1% of 0.96. Additionally, XGB-GRID excelled in binding affinity prediction, with Pearson and Spearman correlation coefficients of 0.87 and 0.73, respectively, showcasing its capability to rank molecules accurately by binding affinity.
SN  - 1420-3049
ER  -