Antimicrobial resistance prediction in Acinetobacter baumannii using collaborative matrix factorization
سال انتشار: 1401
نوع سند: مقاله کنفرانسی
زبان: انگلیسی
مشاهده: 148
نسخه کامل این مقاله ارائه نشده است و در دسترس نمی باشد
- صدور گواهی نمایه سازی
- من نویسنده این مقاله هستم
استخراج به نرم افزارهای پژوهشی:
شناسه ملی سند علمی:
IBIS11_011
تاریخ نمایه سازی: 19 آذر 1402
چکیده مقاله:
Antimicrobial resistance (AMR) is a phenomenon which enables bacteria to survive against antibiotics and becomeresistant to them. AMR is one the most serious crises of the ۲۱st century and must be controlled quickly. In order tocontrol this crisis, it is necessary to determine AMR phenotype of a bacterium to a specific antibiotic. It is possible toanswer this question through laboratory methods, but it is often cheaper and faster to use computational methods.Methods:In this article we proposed a collaborative matrix factorization (CMF) model to predict AMR phenotype of۸۴۵ different strains of A. baumannii bacteria to ۱۲ di↵erent antibiotics. Basic assumption of CMF modelis that there exists a low-dimensional representation of bacteria and antibiotics which makes it possible tomodel AMR phenotype accurately. The purpose of CMF model is to find that d-dimensional latent fea ture space and map both bacteria and antibiotics to this space. To predict AMR phenotype of a pair,CMF model uses inner product of their latent feature vectors. As a feature vector to represent each bac terium, a binary vector of gene’s presence/absence pattern, and for the case of each antibiotic, fingerprintrepresentation was considered. We have used three matrices to create this CMF model; matrix of known bacterium antibiotic phenotypes (phenotype matrix), bacteria similarity matrix and antibiotic similarity matrix. In orderto create bacteria (antibiotic) similarity matrix, similarity of a pair of bacteria (antibiotics) is estimated bysimilarity of their corresponding feature vectors. Finally, proposed model outputs a predicted phenotype matrix.Conclusion:The proposed CMF model predicted a phenotype matrix, which determines AMR phenotype of each strainto each of ۱۲ drugs. The resulting model was evaluated using ۵-fold cross validation and achieved ۸۱.۵%accuracy, ۸۷.۳% sensitivity and ۸۰% area under ROC curve (all in terms of mean cross-validation scores).
کلیدواژه ها:
نویسندگان
zahra seraj
Amirkabir university of technology
fatemeh zare-mirakabad
Amirkabir university of technology.