Adera2.0

Adera2.0 applies deep learning to prioritize drug repurposing candidates for neuroimmunological diseases such as Alzheimer's disease, Parkinson's disease, multiple sclerosis, and depression.

Key Features:

• Three-network architecture: Adera2.0 uses a trio of deep learning networks where the first network encodes textual data from scientific literature into matrix embeddings, the second network predicts outcomes in the form of these embedded matrices using a mean squared error (MSE) loss function, and the third network uses an MSE loss function to extract chemical compound names from relevant sentences.
• Enhanced text mining from PubMed: The workflow automates extraction of chemical compound names and relevant sentences directly from PubMed articles, improving on capabilities from Adera1.0.
• Optimized neural design: Eight different neural network designs were tested and the selected configuration employs Recurrent Neural Networks (RNN) and leaky Rectified Linear Units (leaky ReLU), achieving a reported loss metric of 0.0001 and a sensitivity of 67%.
• Database validation: Predictive outputs have been validated against the DRUG Repurposing Hub database.

Scientific Applications:

• Neuroimmunological drug repurposing: Prioritizes existing drugs for potential therapeutic application in Alzheimer's disease, Parkinson's disease, multiple sclerosis, and depression.
• Literature-driven compound identification: Automates identification and extraction of chemical compound names and relevant sentences from PubMed to support candidate discovery.
• Candidate validation and prioritization: Enables prioritization of repurposing candidates with validation against the DRUG Repurposing Hub database.

Methodology:

Textual data from PubMed are embedded into matrices by an encoder network; a second network predicts embedded matrices using MSE loss; a third network uses MSE loss to extract chemical compound names; eight network designs were compared, and the chosen model uses RNNs with leaky ReLU and was validated against the DRUG Repurposing Hub database.