Junseok Choe

Computer science Ph.D cancidate in Korea Univ.
B. S. Biomedical Science and Chemistry in Korea Univ.

[Junseok Choe @juns941]

If you don't know why the little improvement is also important and how AI can help drug discovery and development which are very time-consuming and cost ineffective, this video is a good supplement.
https://t.co/2l2LDu0yvJ

[Junseok Choe @juns941]

Hi, I am always interested in discovering how drug effects are complex in human body using AI and I found out an attractive paper 'Distributed Representatons of Graphs for Drug Pair Scoring'.

https://t.co/hwyNP6Htwx
#LoG #LearningOnGraphs #GraphPaper

[Junseok Choe @juns941]

There are some margins to improve by using other subgraph pattern induction algorithm. At last, the creating process is individually applicable to many drug pair scoring tasks then may help SOTA models for those tasks. Dataset and model, just we need.

[Junseok Choe @juns941]

Additionally, drug-drug interaction (ddi) prediction is very helpful to novel drug discovery and drug repurpose withoutu time, cost-consuming experiments. If it is not significant quantitavely, even the little improvement can reduce the cost and time drastically.

[Junseok Choe @juns941]

[Potential Impact of DrugPairScoringDR]
DrugPairScoringDR is very useful from various perspectives. It can save drug usage by discovering drug synergy, prevent drug abuse and side effects which cause serious social, cost problems.

[Junseok Choe @juns941]

Here are the paper and poster link for those interested.
Paper: https://t.co/51T7iSSYjx
Poster: https://t.co/UqiF8zEH04
Github: https://t.co/nOjVGDNsiN

[Junseok Choe @juns941]

The authors argue that the augmentation effect of the drug embeddings is significant. Thus, they show the simple MLP only learned the drug representations is better than SOTA models and these SOTA models improve via utilizing the drug representations.

[Junseok Choe @juns941]

Drug Subgraphs are created by WL test for rooted subgraphs and Floyd-Warshall Algorithm for shortest paths. Subsequently, the corpus are optimized to cover novel, induced subgraphs with negative sampling via Skipgram and drug embeddings are optimized simultaneously.

[Junseok Choe @juns941]

As the size and complexity of drug graph sets and graph substructures have increased, prior works have difficulty in dealing with the substructure pattern frequency of drugs. To address it, the authors construct a corpus of the target-context (graph-subgraph) relationship.

[Junseok Choe @juns941]

'DrugPairScoringDR' which may firstly learns and utilizes distributed representations of drugs as additional features and augments SOTA methods for drug pair scoring tasks such as drug synergy, polypharmacy side effects, and ddi prediction.

[Junseok Choe @juns941]

@moonrim_ko
안녕하세요. 일반판 주문 다시 열리길 기다리고 있습니다. 혹시 스위치 없는 모델 판매는 안 하시나요? 스위치 여러 가지를 바꿔가면서 사용할 거라서요