Welcome (back) to Stat 494: Statistical Genetics!

While we wait to get started...

• Sit anywhere you want (just avoid the back two corner tables, please!)
• Catch up on recent messages on Slack
• Please respond to my career exploration poll!
• Reminder: attend at least two department seminars and complete the Speaker Reflection form by the end of the semester
• Turn in your Capstone Reflection
• Review your notes and prepare for today's Journal Club discussion

Goals for today

• Journal Club 6: application of ancestry inference to 23andMe data
• Student presentation
• Instructor debrief
• Project "speed dating"
• Lab 4:
• Part 1 debrief
• Part 2 work time
• If time: Part 2 debrief
• Connecting ancestry inference / PCA to questions of causal inference (confounding, colliders, DAGs)

Journal Club #6

Topic: Genetic Ancestry Inference

• Application of genetic ancestry inference to individuals in the United States
• An early paper using 23andMe customer data (see full publication list here)
• One of the first papers we've seen that focuses on admixed individuals

​

​

Discussion Leaders: Nick, Ronan, Sam

Journal Club Debrief

Submit and upvote questions here →

​

​

[Presenter link for Kelsey]

Stretch Break!

Journal Club Debrief

Key Points:

• SVMs are a useful tool for inferring genetic ancestry similarity
• See ISLR for more details
• Genetic ancestry is highly variable
• Within self-reported race/ethnicity groups
• Across the US
• → can't predict race/ethnicity from genetic ancestry
• … and complex
• → simple models (e.g., assuming K = 2 for African Americans) may not suffice

Figure 5B. The proportion of individuals that self-report as European American, Latino, and African American for each 2% bin of African ancestry and Native American ancestry. The proportion for each 2% bin is shown as a pie chart, with slices colored in proportion to the absolute numbers of individuals from each self-reported identity that carry those levels of genome-wide ancestry. Pie charts are omitted for bins where there were no individuals with those corresponding levels of Native American and African ancestry.

Journal Club Debrief

Connections to previous topics:

• Discussion and justification of choice of population descriptors
• Expanding the focus beyond Europe:
• African Americans, Latinos, European Americans (acknowledging admixture)
• Sub-continental ancestry
• However…
• groups still analyzed separately
• individuals who self-identify as mixed race excluded
• accuracy of statistical methods impacted by data availability (or lack thereof)

p. 38 (Methods)

p. 39 (Methods)

Statistical methods for inferring genetic ancestry

• Machine learning tools (clustering + classification)
• PCA (e.g., smartpca, SNPRelate)
• Regression forests (e.g., RFMix)
• Support vector machines (e.g., 23andMe)
• Model-based, biologically-informed approaches
• Bayesian methods (e.g., STRUCTURE)
• Maximum likelihood estimation (e.g., ADMIXTURE)
• Hidden Markov Models (e.g., STRUCTURE 2.0)
• Li and Stephens population genetic model (e.g., HAPMIX, Chromopainter, FLARE)

Supervised (requires labels or outcome values) → need to have some people with known ancestry so we can train model

Unsupervised (doesn't require labels or outcome values)

Journal Club Debrief - What's Next?

Journal Club #7:

• Date: March 27 (Thursday after spring break)
• Readings: review of statistical methods to handle "population structure" (the definition of which is now expanded to exclude "cryptic relatedness")
• Leaders: Bowman, Cameron, Charles
• Instructions: see the course website
• Remember to share slides with me (leaders) and submit at least one question to slido (everyone else) before class

​

What's Next?

Upcoming Project Checkpoints

Due Dates:

• Checkpoint 1 (Project Preferences Survey): this Friday, March 7
• Checkpoint 2 (Topic Proposal): next Friday, March 14

​

To help with this, let's do a quick round of project "speed dating"...

Project "Speed Dating" - Round 1

Discuss:

1. Introductions
2. What you're hoping to get out of the project (skills, concepts, etc.)
3. Any specific topic ideas you have

Project "Speed Dating" - Round 2

Move to another table.

Make sure there are at least 2 people you didn't talk to during the last round.

Discuss:

• Introductions
• What you're hoping to get out of the project (skills, concepts, etc.)
• Any specific topic ideas you have

Project "Speed Dating" - Round 3

We'll do this one more time in class on Thursday.

(And then I'll give you time in class to complete the Project Preferences Survey.)

​

When you get to class on Thursday, sit with at least 2 people you didn't talk to today!

What's Next?

• Due this Friday (March 7):
• Project Preferences Survey
• Lab 4
• We will not meet as a class next week! Instead:
• Meet with me one-on-one (Monday–Thursday)
• Sign up for a time here: https://calendar.app.google/wFNJ9MgbFPfKToSM9
• Submit your Midterm Learning Reflection at least 48 hours prior
• Meet with your project group to work on the Topic Proposal (due Friday, March 14)
• I'll let you know your groups no later than Monday, March 10
• Work on, and submit, Content Summary 2 (due Friday, March 14)

Understanding PCA

Lab 4: check-in

Part 1:

• How did you calculate MAF?
• HINT: if your results are not between 0 and 1, something needs updating!
• Which SNPs differ in frequency across populations?
• Which SNPs contribute most highly to PC1?
• How many PCs do we need to capture "population structure" in this sample?
• Which plots help us answer this question?

Lab 4: work time

Continue working on Parts 2 and 3!

​

NOTE:

• We'll discuss Omitted Variable Bias together in class on Thursday.
• You can skip this section for now (and leave it blank when you submit).

Challenge: genetic data are observational

Genetic ancestry is a potential confounding variable in GWAS:

allele frequency of the SNP we're testing differs across ancestral populations

environmental factors or causal SNPs elsewhere in genome that differ across ancestry groups

Challenge: genetic data are observational

​

If we know that genetic ancestry is a potential confounding variable, then we should adjust for it in our GWAS models:

E[y | x_j, 𝛑] = 𝛼 + 𝛽_j x_j + 𝛄𝛑,

• y is the trait
• x_j is the number of minor alleles at position j
• 𝛑 is the genetic ancestry
• repeat for all positions j = 1, …, p

← inferred using PCA or other techniques!