Cuckoo Search-Driven Optimization of Artificial Neural Networks for Accurate Fingerprint-Based Toxicity Prediction

Human exposure to a wide range of chemical compounds, some of which pose significant health risks, underscores the critical need to assess chemical toxicity comprehensively. Accurate toxicity assessment is pivotal for minimizing exposure to hazardous substances commonly found in everyday products. High-Throughput Screening (HTS) has traditionally been the go-to method for evaluating toxicity in large-scale chemical assessments. However, HTS has drawbacks, including time-intensive procedures and substantial research costs. Recognizing the limitations of HTS, this study explores alternative approaches to toxicity prediction. Specifically, we investigate the implementation of machine learning techniques to address the shortcomings of HTS. Our research focuses on predicting toxicity utilizing fingerprint datasets and harnessing the power of Artificial Neural Networks (ANNs), which are optimized through the innovative Cuckoo Search Algorithm (CSA). In our study, we identified optimal model configurations, highlighting the effectiveness of ANNs trained with three hidden layers and specific hidden node configurations [61, 69, 105]. Additionally, we employed the Rectified Linear Unit (ReLU) activation function. Our findings demonstrate promising performance, with an F1-Score of 0.6153 and an accuracy of 0.9652, showcasing the potential of this approach in toxicity prediction. This research opens the door to more efficient and cost-effective methods for evaluating chemical toxicity, offering a valuable alternative to conventional HTS techniques.