mhc-peptides-prediction

A set of different models, that predict the degree of binding of peptide proteins.

Short description of the dataset, that was used (https://github.com/ditekunov/mhc-peptides-dataset):

Feature	Description	Type	Example
mhc	Antigen Peptide type	String	HLAA0101
sequence	Letter code of a nucleotide	String	AALEGLSGF
meas	Measurement of a degree of binding	Numeric	0.21281259490425353
pep_class	Whether a peptide is binded or not	Boolean	0

Outcome meas-based models metrics:

Model name	Accuracy real	AUC-ROC real	F-measure
KNN classifier with 1 NN	99.8862 %	-	99.9998 %
Bootstrapped decision tree	99.8863 %	99.9998 %	99.7504 %
Decision tree	99.8863 %	99.9264 %	99.7504 %
Logistic regression	99.8936 %	99.9311 %	99.7664 %
Random forest with 80 trees	99.8863 %	99.9264 %	99.7504 %
Linear regression	80.7395 %	-	99.7304 %

Outcome sequence/mhc-based models metrics:

Model name	Accuracy real	AUC-ROC real	F-measure
Bagged decision tree	85.8888 %	89.9404 %	66.2836 %
KNN classifier with 3 NN	78.9030 %	-	72.1252 %
KNN classifier with 2 NN	80.2825 %	-	67.8703 %
Random forest with 6000 trees	84.5680 %	73.4058 %	60.9688 %
Random forest with 900 trees	84.5570 %	73.3703 %	60.9156 %
Decision tree	81.9189 %	74.7383 %	60.8018 %
Logistic regression	77.2152 %	58.4311 %	-
KNN classifier with 10 NN	78.6461 %	-	38.5544 %

Used libraries:

• numpy
• pandas
• matplotlib
• cv2
• seaborn

Created for HSE University as a course work.

(c) Daniil Tekunov, 2018