Novel VAE Pipeline for Tau Inhibitor Screening in Drosophila Alzheimer's Models

Overview

Alzheimer's disease (AD), a progressive neurodegenerative disorder affecting over 50 million globally, has seen limited success with FDA-approved treatments targeting amyloid beta (Aβ) aggregation, as shown by repeated negative results in recent phase 3 trials. Tau pathology has been shown to be more strongly associated with dementia than Aβ, but research into the mechanisms of Tau is more complex and remains in its initial stages. This study introduces a novel approach integrating computational predictions with validations in vivo to identify therapeutic molecules against Tau aggregation. SMILES representations of molecules known to inhibit Tau are processed through a two-part neural network architecture. The first component, a variational autoencoder (VAE) with recurrent neural network (RNN) encoder and decoder layers, generates novel molecular configurations. The second component, an LSTM-based model, predicts the molecular properties critical to inhibition efficacy. Experimentation involved replacing four vanilla GRU layers in both the encoder and decoder with four bidirectional LSTM and GRU layers, respectively, to evaluate which configuration best suited the data. The output with all predicting properties lying in a predetermined threshold (QED<0.6, SAS<3, 1<logP<5) was Methylene Blue (MB), with validation loss sub-0.2 post 80 epochs. MB has demonstrated efficacy in inhibiting Tau aggregation by accelerating liquid-liquid phase separation, where a homogeneous solution separates into two distinct liquid phases. This anti-aggregatory effect of MB was validated in vivo using Tau-mutant D. Melanogaster. RNA extraction and quantification via spectrophotometry revealed a 10.28% increase in RNA concentration (ng/µL) in the brains of Tau-mutated flies, suggesting a decline in transcriptional suppression. This study underscores the role of computational drug discovery in advancing clinical research, particularly in complex neurodegenerative diseases like Alzheimer's, where traditional empirical methods fall behind.