Scholarlink

<p>This work introduces a deep learning-driven approach focused on the development and implementation of a long short-term memory (LSTM) network for the generation of novel drug-like molecules against hepatocellular carcinoma (HCC). Trained on a curated dataset of FDA-approved drug molecules and accessible sources like ChEMBL, and DrugBank, the model was engineered to learn the sequential patterns in SMILES representations and produce chemically valid, diverse molecular structures with potential therapeutic relevance. The model architecture included an embedding layer of 128 units, a single LSTM layer with 256 hidden units, and a dense output layer with SoftMax activation. Training was conducted over 100 epochs using the Adam optimizer, and validation accuracy and loss curves were monitored to ensure convergence. A total of 100 unique SMILES were generated and subjected to rigorous filtering based on drug-likeness, physico-chemical properties, and synthetic accessibility using cheminformatics tools such as RDKit. To assess their therapeutic potential, the ML-generated compounds were evaluated through in silico validation methods. ADMET profiling using SwissADME, ADMETlab and pkCSM suggested that selected molecules possessed good pharmacokinetic properties and low toxicity risk. Molecular docking against the HCC-relevant protein targets was conducted using the Schrodinger Maestro suite, and favourable binding interactions were observed. Prime MM-GBSA was employed to evaluate the binding free energies of the docked compounds. Two lead structures were advanced to molecular dynamics simulations in Desmond, which confirmed stable binding interactions and favourable binding free energies over a 100 ns simulation period. Overall, the integration of generative deep learning and computational screening techniques in this study demonstrates a viable pipeline for identifying new chemical entities with potential efficacy against HCC. The results provide a foundation for future experimental investigations and further refinement of data-driven drug design methods.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep learning driven de novo smiles generation and in silico validation of potential therapeutics for hepatocellular carcinoma

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月9日

查看全文
  • Mahasin Khan,
  • Judy Jays

摘要

This work introduces a deep learning-driven approach focused on the development and implementation of a long short-term memory (LSTM) network for the generation of novel drug-like molecules against hepatocellular carcinoma (HCC). Trained on a curated dataset of FDA-approved drug molecules and accessible sources like ChEMBL, and DrugBank, the model was engineered to learn the sequential patterns in SMILES representations and produce chemically valid, diverse molecular structures with potential therapeutic relevance. The model architecture included an embedding layer of 128 units, a single LSTM layer with 256 hidden units, and a dense output layer with SoftMax activation. Training was conducted over 100 epochs using the Adam optimizer, and validation accuracy and loss curves were monitored to ensure convergence. A total of 100 unique SMILES were generated and subjected to rigorous filtering based on drug-likeness, physico-chemical properties, and synthetic accessibility using cheminformatics tools such as RDKit. To assess their therapeutic potential, the ML-generated compounds were evaluated through in silico validation methods. ADMET profiling using SwissADME, ADMETlab and pkCSM suggested that selected molecules possessed good pharmacokinetic properties and low toxicity risk. Molecular docking against the HCC-relevant protein targets was conducted using the Schrodinger Maestro suite, and favourable binding interactions were observed. Prime MM-GBSA was employed to evaluate the binding free energies of the docked compounds. Two lead structures were advanced to molecular dynamics simulations in Desmond, which confirmed stable binding interactions and favourable binding free energies over a 100 ns simulation period. Overall, the integration of generative deep learning and computational screening techniques in this study demonstrates a viable pipeline for identifying new chemical entities with potential efficacy against HCC. The results provide a foundation for future experimental investigations and further refinement of data-driven drug design methods.