Golden eagle optimized CONV-LSTM and non-negativity-constrained autoencoder to support spatial and temporal features...

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in drug discovery

Background:

• Drug Response Prediction (DRP) utilizes machine learning (ML) and deep learning (DL) with genomic data.
• DL models excel at feature learning but often ignore prior biological knowledge (e.g., pathway data) due to noisy, multidimensional datasets.
• Noise and class imbalance in DRP datasets reduce accuracy, increase prediction time, and limit applicability.

Purpose of the Study:

• To address noise and class imbalance in DRP datasets.
• To enhance feature learning and extraction for improved drug response prediction.
• To develop a robust hybrid deep learning model for DRP.

Main Methods:

• Application of Non-Negativity-Constrained Auto Encoder (NNCAE) for noise reduction and class balancing.
• Integration of Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) into a hybrid classifier.
• Parameter optimization of the GEO-Conv-LSTM model using the Golden Eagle Optimization (GEO) algorithm.

Main Results:

• The NNCAE-GEO-Conv-LSTM approach achieved high prediction accuracies of 96.99% and 97.79% on two large GDSC datasets.
• Demonstrated significant reduction in processing time and error rates compared to existing methods.
• Successfully learned vital hidden features from pre-processed, balanced, and noise-removed data.

Conclusions:

• The proposed NNCAE-GEO-Conv-LSTM model offers a powerful and efficient solution for drug response prediction.
• Effective handling of data challenges (noise, imbalance) is crucial for accurate DRP.
• This hybrid deep learning approach shows great promise for advancing personalized medicine through accurate drug efficacy prediction.