Developing an intuitive causal graphical model GUI for students and junior researchers

Name: Jonah King
MSc Program: Computer Science MSc

Papers:

An Overview of the Methodologies of Causal Discovery:

Causal discovery: extracts causal direction between variables using data using either constraint based or score based methods
Structural causal models (SCM) are made of three things

1. Exogenous variables: determined by mechanisms extrinsic to the system interested in
2. Endogenous variables: determined by variables intrinsic to the system
3. Structural equations: relates value of target variable X with values of those with directed edges that end at X

Three possibilities in a causal system

1. X causes Y
2. Y directly or indirectly causes X
3. There is a common cause, Z, of both X and Y

Methods and Tools for Causal Discovery and Causal Inference:

Datasets:

LUCAS (lung cancer simple set)

• synthetic dataset with 12 binary variables, 2000 instances and 12 causal relationships
• uses potential factors to develop lung cancer and other related factors
• test if causal discovery methods recover the true causal structure

SIDO (simple drug operation mechanism)

• dataset with 4932 variables adn 12,678 instances
• used to discover the causes for molecular activity in the descriotors

Sachs

• represents proteins and phospholipids present in human immune system cells
• 11 discrete variables, 5400 instances, 17 causal relationships
• aim to discovery connections between the molecules without needing physical intervention on them in a lab