Materials for the Hands-on session

"Machine Learning Algorithm Applications using WEKA"

Downloading WEKA

https://waikato.github.io/weka-wiki/downloading_weka/

Downloading in Windows

Follow the steps here

Basic machine learning theory

The theory is covered in the workshop slides: workshop-Presentation.pdf

Two machine learning applications are demonstrated

1. Supervised ML problem: Binary Classification

• Prediction of drug resistance class from mutation data
• Follow Classification Tutorial
• Datasets: Train · Test

2. Supervised ML problem: Regression

• Prediction of age from DNA methylation
• Follow Regression Tutorial
• Datasets: Train · Test · Labelled Test

Data description

Classification (Classification_train.csv, Classification_test.csv)

Each row is one sample. There are 54 columns, and the files have no ID column.

Column	Description
CATEGORY	Target class (nominal): S or M. S is the susceptible class and M is the resistant class.
53 mutation columns, e.g. A1673558C, G7362C, TGA1674966GGA	Numeric features, one per mutation. The name encodes the change: reference base(s), then genome position, then the alternate base(s). For example, A1673558C is A at position 1673558 changed to C. A value of 0 means the mutation is absent. Values close to 1 mean it is present in nearly all reads.

• Train: 342 samples (203 M, 139 S). Test: 84 samples (50 M, 34 S).
• The test file includes the CATEGORY column, so you can check your predictions against it.

Regression (Regression_train.csv, Regression_test.csv, Regression_test_labelled.csv)

Each row is one sample. There are 14 columns.

Column	Description
RPA2_3, ZYG11A_4, F5_2, HOXC4_1, NKIRAS2_2, MEIS1_1, SAMD10_2, GRM2_9, TRIM59_5, LDB2_3, ELOVL2_6, DDO_1, KLF14_2	Numeric features, 13 in total. These are DNA methylation markers. The name is the gene symbol plus a numeric suffix that identifies the measured site within that gene. Values are on a 0-100 scale, probably methylation percentage. MEIS1_1 corresponds to MEIS1-AS3 in the original paper.
Age	Target (numeric): age in years. The training range is 18-69.

• Train: 208 samples. Test: 104 samples.
• Data source: The dataset appears to come from the forensic age-prediction study Chronological age prediction based on DNA methylation: Massive parallel sequencing and random forest regression (Forensic Science International: Genetics, 2017). The sample counts (208 train, 104 test) and age range (18-69) match that study. It reports MAD 3.16 and RMSE 3.93 years on its test set, which you can compare with your WEKA results.
• Regression_test.csv has the Age column empty, so you can predict it in WEKA.
• Regression_test_labelled.csv is the same file with the true ages filled in, so you can check your predictions.

License

These materials are licensed under CC BY 4.0. You are free to share and adapt them with attribution to Abhirupa Ghosh.