About - SDLDpred

SDLDpred is a web-based tool for prediction of drugs of lifestyle-related diseases from symptom data.

It used unsupervised machine learning models trained with Bisecting K-Means algorithm using drug-symptom association data of 143 lifestyle-related diseases, with focus on pulmonary and cardiovascular diseases. A summary of the diseases are given in the following table.

Disease category	Number of diseases
Pulmonary diseases	67
Cardiovascular diseases	56
Musculoskeletal diseases	15
Diabetes	4
Obesity	1
Total	143

List of diseases: Download [TSV file]; List of drugs: Download [TSV file]; List of symptoms: Download [TSV file]

Association datasets:

Association	Data source	Description	Download
Drug-disease associations	Comparative Toxicogenomics Database (CTD)	3159 associations; 1271 drugs; 143 diseases	Matrix (TSV file) Download Edge-list (TSV file) Download
Symptom-disease associations	Human Symptom Disease Network (HSDN)	9466 associations; 305 symptoms; 143 diseases	Matrix (TSV file) Download Edge-list (TSV file) Download

The drug-symptom associations were computed based on (i) cosine similarity ( A_{drug-symptom<cosine>} ), (ii) Pearson's Correlation Coefficient ( A_{drug-symptom<pearson>} ), and (iii) Jacard similarity ( A_{drug-symptom<jaccard>} ) between each drug and symptom represented as a vector of disease-associations.

dr_i = i^th drug represented as a vector of associations with the diseases, s_j = j^th symptom represented as a vector of associations with the diseases, ||x|| = Magnitude of the vector x.

The drug-symptom association dataset:

A_{drug-symptom<cosine>}	Download [Matrix (TSV file)] Download [Edge-list (TSV file)]
A_{drug-symptom<pearson>}	Download [Matrix (TSV file)] Download [Edge-list (TSV file)]
A_{drug-symptom<jaccard>}	Download [Matrix (TSV file)] Download [Edge-list (TSV file)]

SDLDpred provides a confidence score for each predicted drug. It is computed using the following equation. D = random variable denoting the pairwise euclidean distances between the drugs; e = euclidean distance of a predicted drug; f_dist = probability density function of the random variable D.

Please contact Dr. Sudipto Saha (ssaha4@jcbose.ac.in, ssaha4@gmail.com) regarding any further queries.