This code presented the work of the bachelor thesis 'Exploring peptide-centric HLA-type prediction with data mining and machine learning'. The three explorative typing approaches (NetMHCpan approach, allele-associated-lookup approach and Random Forest approach) can be evaluated on the TueDB. The results of all approaches are in a result txt in the corresponding folder. If you want to try the peptide-centric HLA typing, the best-performing allele-associated lookup approach can be executed in the allele_associated_lookup_allele_specific_peptides.ipynb.

0 - Baseline

The baseline for 2-digit and 4-digit prediction can be run in the baseline.ipynb.

Evaluation possible on:

• 2-digit
• 4-digit

1 - Ghosh Typing

The internal standard 2-digit prediction can be run with ghosh_typing.py.

2 - NetMHCpan-approach

The NetMHCpan approach can be run and evaluated in the netmhcpan_run_evaluation.ipynb in three variations.

• Option 1
• Option 2
• Option 3

Evaluation possible on:

• 2-digit
• 4-digit

3 - Allele-associated-lookup approach

The allele-associated lookup approach has a run and an evaluation file. The run can be performed in five variations.

• Option 1: run allele_associated_lookup_all_peptides.ipynb to get a result_dict, run allele_associated_lookup_evaluation.ipynb
• Option 2: run allele_associated_lookup_all_peptides.ipynb to get a result_dict, run allele_associated_lookup_evaluation.ipynb
• Option 3: run allele_associated_lookup_all_peptides.ipynb to get a result_dict, run allele_associated_lookup_evaluation.ipynb
• Option 4: run allele_associated_lookup_allele_specific_peptides.ipynb to get a result_dict, run allele_associated_lookup_evaluation.ipynb
• Option 5: run allele_associated_lookup_allele_specific_peptides.ipynb to get a result_dict, run allele_associated_lookup_evaluation.ipynb (the gibbs clustered dataset can be generated in the gibbs-clustering folder)

Evaluation possible on:

• 2-digit
• 4-digit

4 - Random-forest-approach

The random forest approach can run with the RandomForest.ipynb

Evaluation possible on:

• 2-digit
• 4-digit

Allele-associated dataset

The allele-associated dataset was generated from data from these mhcquant commands.

TueDB

The TueDB was preprocessed and adapted to the use case.

DATA

The DATA folder comprises 4.62 GB and entails the TueDB, allele-associated-dataset and additional external data. The DATA folder can be requested on demand.