BMSC-GA 4438: Computational Causal Discovery
New York University
Spring 2014
Computational Causal Discovery, 3 credits
Course meeting times and location:
•
•
227 East 30th Street (Translational Research Building), TBD
Friday afternoon (TBD)
Faculty contact info & office location:
Alexander Statnikov, Ph.D.
• 227 East 30th Street (Translational Research Building), 7th floor, Office #736
• Email: [email protected]
• Office phone: 212-263-3641
• Cell phone: 615-545-3685
Constantin F. Aliferis, M.D., Ph.D., F.A.C.M.I.
• 227 East 30th Street (Translational Research Building), 7th floor, Office #743
• Email: [email protected]
• Office phone: 212-263-5281
Prerequisites: Foundational coursework in Mathematics, Statistics, and Computer Science is required.
Specifically, students are required to take the following courses prior to registering: Calculus (2
semesters), Linear Algebra, Algorithms and Data Structures, Introduction to Programming, Probability
and Statistics. In some circumstances, students who do not meet the above requirements can obtain
permission of the instructor to take the course.
Course topics by weeks:
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
Introduction to Computational Causal Discovery
Causal Discovery from Randomized Data: Essential Epidemiology and Randomized Trials
Concept of Causation
Association and Independence
From Causation to Association
From Association to Causation
Overview of Formal Preliminaries, Essential Causation Axioms and Explications
Causal Discovery from Observational Data w/o Hidden Variables: SGS & PC algorithms (week 1)
Causal Discovery from Observational Data w/o Hidden Variables: SGS & PC algorithms (week 2)
Causal Discovery from Observational Data w/o Hidden Variables: GLL algorithms. Connection
between Causation, Prediction, and Feature Selection
11. Causal Discovery from Observational Data w/o Hidden Variables: Extensions to GLL algorithms
and LGL algorithms
12. De-Novo Reverse Engineering of Gene Networks
13. Causal Orientation Techniques
14. Causal Discovery from Observational Data w/o Hidden Variables: Modified PC, FCI & IC*
algorithms
15. Using Computational Causal Discovery to Design Efficient and Accurate Experiments
Course materials:
Required books:
1) Spirtes,P., Glymour,C.N. and Scheines,R. Causation, prediction, and search (2nd edition). MIT
Press, Cambridge, Massachusetts: 2000. Electronic version is available from:
http://cognet.mit.edu/library/books/view?isbn=0262194406.
2) Attaway S. Matlab, Second Edition: A Practical Introduction to Programming and Problem
Solving (Second edition). Butterworth-Heinemann: 2011.
Optional book:
3) Pearl,J. (2009) Causality: models, reasoning, and inference. Cambridge University Press,
Cambridge, U.K.
Online course from CMU:
https://oli.cmu.edu/jcourse/webui/guest/join.do?section=csr
Computer tools and software:
•
•
•
•
Matlab. Required toolboxes: Bioinformatics, Statistics, Neural Network, Optimization.
Causal Explorer library for Matlab (http://www.dsl-lab.org/causal_explorer/)
TETRAD (http://www.phil.cmu.edu/projects/tetrad/current.html)
Causality Lab (http://www.phil.cmu.edu/projects/causality-lab/)
Learning objectives: Understand methods and techniques for causal structure discovery from
randomized and observational data.
Assignments and assessment:
•
•
•
•
•
•
•
Readings assignments (textbook, articles)
Practical take-home assignment #1: Intro to Matlab
Practical take-home assignment #2: Evaluation of methods for causal discovery
Practical take-home assignment #3: Discovery of gene regulatory networks from gene
expression data
Weekly homework assignments geared towards understanding of algorithms
Midterm examination
Final examination
Each practical take-home assignment contributes 10% of the grade (30% for practical assignments in
total), each examination contributes 30% of the grade (60% for exams in total), homework assignments
contribute a total of 10%.
Course Assessment: This course can be taken for a letter grade or audited.