K-mixdd Documentation

Description
K-medoids algorithm with an augmented center initiation method. The method is based on a mixed cost function, which involves the multiplication of the kernel density of the principal components of the datasets and the minimal distance of the data object to its nearest centers.

Homepage
http://www.thua45.cn/grit-2.0

Release
Source code 1.0      01/12/2025    http://www.thua45.cn/grit-2.0/programs/k-mixdd-1.0/k-mixdd-1.0.zip

Install
Go the source folder and run "g++ *.cpp -std=c++11 -lpthread -o k-mixdd" command, a binary named "k-mixdd" will produced in the folder.
Under Windows OS try "g++ *.cpp -std=c++11 -static -lpthread -o k-mixdd.exe" instead.

Requirement
Minimum 8GB RAM, if you install from the source code, the g++ complier is also required.

Usage
k-mixdd -i input -k cluster [-c cycles] [-r sampling_n] [-m ini_method] [-a alpha] [-d distance] -o output

Options
-i input matrix.
-k number of clusters.
-c max number of iteration cycles.
-r sub-sampling.
-m center initialization method. 0 for k-mixdd, 1 for k-means++
-a contribution parameter for density and distance, usually between 0.5 and 2.0.
-d distance, 0 for Euclidean distance, 1 for Pearson correlation, 2 for Jaccard distance, 3 for Spearman correlation, 4 for Cosine similarity.
-o output, output file name.

Example Data
input matrix: Homo_sapiens_FLAVER.matrix.Label-ID	http://www.thua45.cn/grit-2.0/programs/k-mixdd-1.0/Homo_sapiens_FLAVER.matrix.zip

Example
An example run should like: k-mixdd-1.0.exe -i Homo_sapiens_FLAVER.matrix.Label-ID -k 7 -d 4 -m 0 -a 2.0 -o kmixdd_d4_m0_a2.0_k7_rAll
This command takes one input file: Homo_sapiens_FLAVER.matrix.Label-ID. It will produce an output file named kmixdd_d4_m0_a2.0_k7_rAll.txt. Full instruction is avaiavle in Flaver's user manual.

References
Tinghua Huang, Min Yao. FLAVER 2.0: ENHANCED PERFORMANCE IN THE IDENTIFICATION OF KEY TRANSCRIPTION FACTORS FROM TRANSCRIPTOME DATA. 2015 Jun.

Contact
Dr. Tinghua Huang, thua45@126.com
Dr. Min Yao, minyao@yangtzeu.edu.cn