deepinsight_py

Python implementation of the DeepInsight algorithm.

Originally it was developed in MATLAB by the following authors, I own no credit.

Original Paper: https://www.nature.com/articles/s41598-019-47765-6 Alok Sharma, Edwin Vans, Daichi Shigemizu, Keith A Boroevich, and Tatsuhiko Tsunoda. Deepinsight: A methodology to transform a non-image data to an image for convolution neural network architecture. Scientific reports, 9(1):1–7, 2019

INPUT: TSNE LOCATIONS

OUTPUT: IMAGE GENERATED BY THE ALGORITHM

Algorithm works as follows:

1: The user has to perform tsne/kpca/umap (any dimensionality reduction algorithm) and generate the locations. I got the following plot when I performed tsne on some gene expression data

2: Find the vertices of the Minimum Bounding Rectangle

3: Rotate the image

4: Convert from cartesian domain to pixel domain Step 1-2-3 shown together

5: Map the respective intensities to the coordinates

Image is ready to be used in any CNN architecture