DR3E-Net: Drug Repurposing Network Framework Based on Deep Embedding, Differentially Expressed Genes, and Drug Side Effects

In contrast to traditional de-novo drug discovery, which involves high costs and lengthy development times alongside a high failure rate, drug repurposing (DR) offers a potential solution to these challenges. This study introduces DR3E-Net, an in-silico DR framework that integrates embedded vectors of genetic expression influenced by drugs, differentially expressed genes (DEGs) associated with diseases, and information about drug side effects. The DR3E-Net utilized LINCS L1000 datasets, which comprised 680,909 samples, 3,820 drugs, and 977 genes after preprocessing. The deep-embedding model within DR3E-Net was trained on this dataset, facilitating the transformation of 977-dimensional vectors into 32-dimensional unit-hypersphere embeddings. Subsequently, cosine similarity (CS) was computed among them (Step 1). To further elaborate, agonistic drugs against DEG linked to a disease (Alzheimers disease (AD) in this study) were merged with outcomes from Step 1 (Step 2). Finally, the SIDER and SAEDR/DRIP databases were combined with the integrated results from Step 2 (Step 3). The DR3E-Net was utilized to predict potential alternative drug candidates for three AD medications: donepezil, galantamine, and memantine. The drugs predicted by the deep-embedding model (Step 1) were combined with agonistic drugs against CRTAP gene, resulting in 111, 86, and 96 drugs, respectively (Step 2). Lastly, drug side effect information was incorporated with the aforementioned drug outcomes, yielding 21, 15, and 18 drugs (Step 3). DR3E-Net was developed to predict functionally similar drugs while considering disease-associated DEGs of patients and drug side effect information.