Denoising autoencoder multilayer perceptron spiking neural network for isonicotinic acid yield prediction on real industrial dataset

Isonicotinic acid (INA) has attracted considerable interest as a crucial pharmaceutical intermediate, especially for the production of the anti-tuberculosis drug isoniazid. Nonetheless, industrial production of INA encompasses intricate procedures that are highly sensitive to process parameters, leading to yield variability. Hence, an efficient prediction model for forecasting INA yield is essential for enhancing production yields and ensuring the consistency of INA in pharmaceutical manufacturing processes. To address this challenge, the present study developed a brain-inspired spiking neural network (SNN) tailored to the prediction of INA yield. Specifically, we propose a novel denoising autoencoder multilayer perceptron based spiking neural network (DAEMLP-SNN) for this purpose. The SNN is designed to accurately emulate the dynamic behavior of biological neurons while maintaining low power consumption, thereby ensuring high biological plausibility. Drawing upon the principles of autoencoders, our research constructs a denoising autoencoder SNN capable of extracting meaningful latent features and compressing high-dimensional industrial data. Moreover, ​we concatenated the extracted features with the original data, thereby creating a more comprehensive representation of the input. This enriched input was then fed into the multilayer perceptron SNN, which markedly enhances the robustness and precision of INA yield predictions. Experimental findings demonstrated the superior performance of DAEMLP-SNN, as it consistently achieved accurate predictions across diverse process parameters.