/deep_omics

Repository for Short Course: Deep Learning in Omics

Primary LanguageHTMLMIT LicenseMIT

Deep Omics

This repository saves some examples used in the Short Course on Deep Learning in Omics, by Wei Sun and Nancy Zhang.

logo

Use auto-encoder to analyze somatic mutation datas

see folder somatic_mutations/msk_impact

scRNA-seq data

see folder scRNAseq. There are two datasets:

  • Allen_BI: single nucleus RNA-seq data of human brains, generated by SMART-seq2 protocol

  • Zheng_2017: scRNA-seq data of human blood, generated by 10x droplet-based system. Cells of known cell types were mixed.

  • gene_annotation: gene annotation data and summary.

hyper_parameter_tuning

Examples showing hyper-parameter tuning when studying the association between HLA and TCR

Terminology

Node/neuron: a computational unit in a neural network that has one or more weighted input connections and an activation function that combines the inputs and provide an output.

Weights: when input enters the neuron, it is multiplied by a weight.

Embedding: low dimensional representation of a high dimensional input

Activation function: the output of a neuron is f(Wh + b), where h is input, W is the weight, b is bias, and f is the activation function.

Perceptron: a perceptron is a simple linear binary classifier. For example, y = 1 if WX + b > 0 and y = 0 otherwise.

Feedforward neural network: a neuron network where information goes forward.

MLP (Multi Layer perceptron): in a narrow sense, it refers to multiple layers of perceptrons, but it often refers to any feedforward neural networks.