Inspiration
In the face of rising drug development expenses and the rising costs of generating new treatments from the ground up, repurposing existing drugs for new purposes is an appealing alternative that can considerably lower safety concerns and development costs. We present a rapid drug development tool based on knowledge graphs that builds embeddings with edge and relation information from the knowledge graph using deep learning methods, and then leverages TigreGraph's fast querying power to locate information about the anticipated components from the target medication, allowing the predictions to be refined further. Drug development specialists will be aided in creating, examining, and comprehending drug repurposing predictions thanks to the visualization and interpretation of the predictions. We may not only further analyze the results based on deep learning using Tigergraph's data query capabilities and the abundant knowledge in the knowledge graph, but also decrease the danger of the projected medicine by performing a hazardous analysis of the predicted outcomes based on the current knowledge. Our findings suggest that the tool is capable of making accurate medication predictions and providing useful information.
What it does
We provide interactive graphic descriptions of biological mechanisms to aid in knowledge-based drug repurposing in this section. To accomplish this, we create a medication repurposing knowledge graph. data on molecular interactions, gene expression, clinical trials, and pharmacological therapies Then we utilize TranSE to train a new term embedding in order to forecast various known drug alternatives based on known valid drug treatments. Finally, biological knowledge about the alternative pharmaceuticals in the knowledge graph of opportunity is used to enlist the help of drug experts in the final drug choice.
How we built it
A Drug Repurposing Knowledge Graph, an engine that can swiftly extract knowledge on the knowledge graph based on drug nodes, and lastly a deep learning embedding approach that can map multi-relational high-dimensional knowledge from the knowledge graph to a low-dimensional space and model it effectively are the three pieces of the process. It's worth mentioning that each of these components is self-contained and may be replaced without affecting the others. The Drug Repurposing Knowledge Graph, for example, is based on the public knowledge graph DrugBank, GNBR, and some knowledge taken from the medical literature, and covers a total of drug components, symptoms, diseases, genes, molecular components, side effects, and other information. As a knowledge graph, we investigate the medication prediction problem. The knowledge graph-based drug prediction problem is viewed as a knowledge graph complementary problem, in which drug prediction is accomplished by predicting whether a link between drug and sickness exists. In addition, we will primarily detail the adverse effects in the interaction phase for the purpose of determining safety. In addition, the anticipated medications will be visualized using embedding to determine their variety, offering some interpretability.
Challenges we ran into
The challenge we're trying to answer is how to apply deep learning for reliable drug development. To achieve this goal, you'll need to solve two major issues: how to integrate multimodal data into the knowledge network, and how to make your forecasts more understandable. We chose TransE as the strategy for embedding high-dimensional information in the first challenge. TranSE is a well-known method for sinking multi-relational knowledge graphs, and we employ the TransE embedding method, which has already been trained. The second issue is that there are many interpretable approaches, but the majority of them are reliant on the model's self-explanation. Here, we leverage visual interactions to entice human scientists to participate in drug development, thereby artificially reducing the risk of drug development.
Accomplishments that we're proud of
We employed COVID-19 as a test case, and after both model center and user center screening, over 60% of the medications on the drug list were successful in treating COVID-19.
What we learned
We discovered that utilizing knowledge graphs to reposition drugs has a lot of promise and might be a wonderful idea for future medication development.
What's next for Deep Drug Repurposing By Involving Visual Interaction
The existing knowledge graph has a long tail of medications, with a lot of data concentrated on a few pharmaceuticals and very little data on the rest, and the process of embedding the knowledge graph has to be improved to offer more accurate data on drug candidates.
Log in or sign up for Devpost to join the conversation.