AI-components

Introduction

AI Components of GeneSourceCode project is a subproject taking charge of machine learning related tasks. AI Components aim to make full use of gene and other medical data with the facilities of modern AI technologies. AI Components currently focus on

1.Prediction of diseases and traits directly from raw DNA sequence. We will test both traditional classification/regression algorithm and popular deep neural network ways such as LSTM, p-LSTM, IndRNN, attention model and so on to process raw DNA sequential data.

2.Prediction of diseases and traits from hand-designed feature. The hand-designed feature extracted from raw DNA, RNA or histone sometime may be discriminative enough to make prediction task viable. We will try to extract and learn on features this way.

3.Medical application based on visual clues. Computer Vision has become a reliable way of prediction after deep learning prevails. Medical scientists have adopted this method to various applications such as predicting or detecting certain diseases, image processing on X ray pictures, and so on. We will implement all these applications in this subproject and make them optional service modules.

4.Mining fitness status on physical examination and motion data, We will also mining data provided by users to detect potential fitness problem or reveal healthy status.

Components

1.Prediction of diseases and traits directly from raw DNA sequence.

(1)rawDNA/LSTM: classification base on DNA subsequence:

You can train a classifier with train_LSTM. The dataset generation tools will be released soon.

2.Prediction of disease and trais from hand-designed feature.

(1)extractedDNA/AllelesClassifier: classification base on Alleles

You can train a classifier on polymorphic alleles.

3.Medical appliation based on visual clues.

(1)visual/facial: classification based on facial images:

(2)visual/iris: biometric identification and illness detection according to visual information from iris.

You can train a classifier with train_facial_classifier. The dataset generation tools will be released soon.

4.Mining fitness status on physical examination and motion data.