Skip to content

Demo 2: Project Planning Approach - Gene Expression in Transplant Patients

Scenario: Your PI just handed you a gene expression dataset from transplant patients with no clear direction, just a deadline. You need to figure out good project options through systematic exploration.

The Dataset Context

You've been given: - Gene expression data from 200 transplant patients - Patient metadata: age, time post-transplant, rejection status, hospital site - 20,000+ genes measured via RNA-seq - A vague directive: "Find something interesting for the grant renewal"

Step 1: PI Interview - Understanding Research Context (5 minutes)

Initial Questions to Ask Your PI

Research Background: - "What has your lab's previous research focused on in transplant medicine?" - "What were you hoping to see from this dataset when it was collected?" - "What would be the best possible outcome from this data for the grant?"

Hypothetical PI Responses:

PI Response 1: "We've been studying rejection mechanisms for 10 years. I want to find biomarkers that can predict rejection before it becomes clinically apparent."

PI Response 2: "This is part of a multi-site study. We're interested in understanding why some patients do better at certain hospitals."

PI Response 3: "We collected this as part of a larger biorepository. I'm open to any direction that could lead to a high-impact publication."

Choose Response 1 for this demo

Follow-up Questions Based on Response 1:

  • "What biomarkers have you looked at before?"
  • "How early can we predict rejection with current methods?"
  • "Are there specific gene pathways you suspect are involved?"
  • "What would make a biomarker clinically useful?"

PI's Elaboration: "Current biopsies can only detect rejection after tissue damage starts. If we could predict it 2-3 months earlier using blood tests, that would be revolutionary. I suspect immune activation pathways are key."

Step 2: Data Exploration - Understanding What You Have (8 minutes)

Hypothetical Data Schema

Patient Metadata (200 patients):
- patient_id: P001, P002, ...
- age: 25-75 years
- time_post_transplant: 0.5-5 years
- rejection_status: No_Rejection (140), Rejection (60)
- hospital_site: Site_A (80), Site_B (70), Site_C (50)
- sample_type: blood, biopsy

Gene Expression (20,000 genes):
- Genes: ENSG00000000001, ENSG00000000002, ...
- Expression levels: Log2(TPM + 1)
- Quality metrics: % mitochondrial genes, total reads

Initial Exploration Questions

  1. Data Quality Assessment:
  2. Are rejection rates similar across sites? (Site_A: 25%, Site_B: 35%, Site_C: 30%)
  3. Do we have batch effects? (Samples processed in 3 batches over 2 years)

  4. Missing data patterns? (5% missing expression values, random distribution)

  5. Clinical Patterns:

  6. Rejection timing: Early (0-1 year): 40%, Late (1+ years): 60%
  7. Age correlation: Older patients slightly higher rejection risk

  8. Site differences: Site_B has more complex cases

  9. Expression Data Quality:

  10. Distribution looks normal after log transformation
  11. Some genes have very low expression across all samples
  12. Potential outlier samples: 3 samples with unusual expression patterns

Preliminary Findings to Discuss with PI

"I found some interesting patterns. Site_B has higher rejection rates - they might be taking sicker patients. Also, we have good temporal coverage for early prediction. Should we focus on early rejection specifically?"

Step 3: Literature Search - Informing Your Understanding (5 minutes)

Key Search Terms Based on Initial Findings

  1. "transplant rejection biomarkers blood RNA-seq"
  2. Recent reviews show most studies use biopsy tissue
  3. Blood-based RNA signatures are emerging field

  4. 2-3 commercial tests exist but limited validation

  5. "early rejection prediction gene expression"

  6. Most studies focus on detecting ongoing rejection
  7. Few studies predict rejection weeks/months ahead

  8. Gap in literature for early warning systems

  9. "immune activation pathways transplant"

  10. JAK-STAT, interferon response pathways well established
  11. T-cell activation markers consistently upregulated
  12. Novel: metabolic reprogramming during rejection

Literature Summary for PI

"Good news - there's a clear gap in early prediction. Most studies detect ongoing rejection, not future risk. The interferon pathway is consistently involved, but metabolic changes are understudied. This could be novel."

Step 4: Hypothesis Refinement with PI (7 minutes)

Present Options to PI

Option 1: Early Rejection Prediction Model - Focus: Predict rejection 2-3 months before clinical detection - Methods: Machine learning with temporal analysis - Timeline: 6 months for initial results - Clinical impact: Could enable preventive interventions

Option 2: Site-Specific Rejection Mechanisms - Focus: Understand why Site_B has higher rejection rates - Methods: Differential expression + pathway analysis - Timeline: 3-4 months - Clinical impact: Improve patient selection/treatment protocols

Option 3: Novel Metabolic Pathway Discovery - Focus: Identify new biological mechanisms - Methods: Network analysis + functional validation - Timeline: 12-18 months - Clinical impact: New therapeutic targets

PI Feedback Session

PI: "Option 1 sounds most fundable, but is 200 patients enough for machine learning?"

Your response: "We could start with a simpler approach - identify the strongest individual biomarkers first, then build complexity. Also check if we can access the larger multi-site dataset you mentioned."

PI: "Good point. And for the grant, we need preliminary data in 3 months. What's the minimum viable analysis?"

Refined Plan: 1. Month 1: Identify top 20 genes associated with future rejection 2. Month 2: Validate in cross-validation, test simple prediction model 3. Month 3: Generate preliminary figures for grant, plan larger validation study

Step 5: Next Investigative Cycle (3 minutes)

Specific Next Steps

  1. Technical Questions:
  2. Should we include patients with borderline rejection?
  3. Do we need to account for immunosuppressive medications?

  4. How do we handle time-varying gene expression?

  5. Resource Planning:

  6. Need bioinformatics support for pathway analysis
  7. Might need access to larger validation cohort

  8. Budget for qPCR validation of top candidates

  9. Validation Strategy:

  10. Internal cross-validation first
  11. External validation in year 2 with new cohort
  12. Prospective validation in years 3-4

Return to PI with Specific Plan

"Based on our discussion, I propose a three-phase approach: discovery (months 1-3), validation (months 4-12), and prospective testing (years 2-3). The first phase will give us grant-quality preliminary data, and we can assess feasibility for the larger study."

Success Criteria

✅ Systematic Approach: Used iterative PI consultation to refine direction ✅ Literature-Informed: Identified specific gaps and opportunities
✅ Feasible Planning: Broke large question into manageable phases ✅ Resource-Aware: Considered timeline, budget, and expertise constraints ✅ Clinical Relevance: Maintained focus on translational impact

Key Takeaways

  • Always start with stakeholder input - PIs have context you don't
  • Literature search early - Don't reinvent the wheel
  • Iterative refinement - Present options, get feedback, adjust
  • Plan in phases - Minimum viable analysis → full study → validation
  • Consider resources - Time, money, expertise all constrain what's possible