Scholarlink

Background <p>The potential of synergistic drug combinations in cancer research has been acknowledged for their ability to enhance treatment outcomes. Instead of depending on costly experimental methods, computational techniques have emerged to forecast such combinations. Recent advancements in computational methods have utilized vast datasets comprising chemical, genomic, and pharmacological information. These methods aim to predict drug interactions by analyzing complex biological data and identifying patterns that are indicative of synergistic effects.</p> Methods <p>Here, SynergyImage employs an unsupervised pre-learning technique, ImageMol, to extract features from the chemical structure images of the drugs. Concurrently, gene expression data is transformed into image formats using the DeepInsight method, enabling the extraction of image-based features corresponding to the cancer cell lines through a convolutional neural network. Following a dimensionality reduction step, the framework utilizes a multi-layer perceptron network to predict the synergy scores for various drug combinations.</p> Results <p>Experimental results show that SynergyImage surpasses leading methods on the O'Neil benchmark datasets, highlighting its capability to predict synergistic drug combinations. In particular, SynergyImage recorded an MSE of 73.402 ± 1.185 and a PCC of 0.83 ± 0.003. These numerical results further highlight the superior performance of SynergyImage compared to other models.</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

SynergyImage: image-based model for drug combinations synergy score prediction

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月21日

查看全文
  • Maryam Mehrabani,
  • Amir Lakizadeh,
  • Alireza Fotuhi Siahpirani,
  • Ali Masoudi-Nejad

摘要

Background

The potential of synergistic drug combinations in cancer research has been acknowledged for their ability to enhance treatment outcomes. Instead of depending on costly experimental methods, computational techniques have emerged to forecast such combinations. Recent advancements in computational methods have utilized vast datasets comprising chemical, genomic, and pharmacological information. These methods aim to predict drug interactions by analyzing complex biological data and identifying patterns that are indicative of synergistic effects.

Methods

Here, SynergyImage employs an unsupervised pre-learning technique, ImageMol, to extract features from the chemical structure images of the drugs. Concurrently, gene expression data is transformed into image formats using the DeepInsight method, enabling the extraction of image-based features corresponding to the cancer cell lines through a convolutional neural network. Following a dimensionality reduction step, the framework utilizes a multi-layer perceptron network to predict the synergy scores for various drug combinations.

Results

Experimental results show that SynergyImage surpasses leading methods on the O'Neil benchmark datasets, highlighting its capability to predict synergistic drug combinations. In particular, SynergyImage recorded an MSE of 73.402 ± 1.185 and a PCC of 0.83 ± 0.003. These numerical results further highlight the superior performance of SynergyImage compared to other models.