RNA Velocity and Trajectory Analysis with ChatSpatial

Learn how to uncover cellular dynamics, developmental trajectories, and temporal processes in your spatial transcriptomics data through natural conversation with ChatSpatial.

What You Will Discover

By the end of this tutorial, you will know how to:

Ask about cell dynamics in your tissue using natural language
Compute RNA velocity to see directional changes in gene expression
Infer developmental trajectories and cell fate transitions
Explore spatial-temporal patterns across tissue regions
Interpret biological significance of velocity and trajectory results
Choose the right method for your specific research question

Before You Start

What You Need

✅ Spatial data loaded in ChatSpatial
✅ Basic preprocessing completed (see Basic Analysis)
✅ Interest in cellular dynamics and development

What Are RNA Velocity & Trajectories?

RNA Velocity reveals the direction and speed of cellular changes by measuring the balance between newly transcribed (unspliced) and mature (spliced) RNA molecules. Think of it as seeing which direction cells are “moving” in gene expression space.

Trajectory Analysis reconstructs developmental paths and cell state transitions, helping you understand how cells differentiate, mature, or respond to stimuli over time.

Biological Background

Understanding RNA Velocity

RNA velocity is based on a simple biological principle:

Genes are transcribed → unspliced pre-mRNA is created
Pre-mRNA is processed → spliced mature mRNA is produced
mRNA is degraded → both forms eventually disappear

The ratio of unspliced to spliced RNA tells us:

High unspliced/spliced: Gene is being actively turned ON
Low unspliced/spliced: Gene is being turned OFF
Balanced ratio: Gene expression is in steady state

Why Spatial RNA Velocity Matters

Traditional RNA velocity analyzes cells in isolation, but spatial velocity reveals:

Where developmental processes occur in tissue
How cellular transitions spread spatially
Whether neighboring cells follow similar trajectories
Which tissue regions drive developmental processes

Step 1: Starting with RNA Velocity Analysis

💬 The Conversation Begins

👤 You: “I have a developing brain tissue dataset and I am curious about cellular dynamics. Can you analyze RNA velocity to see how cells are changing over time?”

🤖 ChatSpatial: “Good question! RNA velocity analysis is good for understanding developmental dynamics in brain tissue. Let me check your data and compute RNA velocity for you.

First, let me examine what type of velocity data you have available…”

[ChatSpatial analyzes data structure]

🤖 ChatSpatial: “I found that your dataset contains both spliced and unspliced RNA counts, which is suitable for direct velocity computation. Here are your options:

Available Velocity Methods:

scVelo (recommended) - Fast and reliable, suitable for most analyses
- Three modes: stochastic (default), deterministic, dynamical
VeloVI - Deep learning approach for velocity computation, more accurate but slower

Note: These methods compute RNA velocity. For trajectory inference, use CellRank, Palantir, or DPT with the computed velocity.

Which velocity method would you prefer, or should I start with scVelo?”

👤 You: “Let us start with scVelo. I want to understand the basics first.”

🤖 ChatSpatial: “Good choice! scVelo is good for getting started. Let me compute RNA velocity using the stochastic mode, which works well for most spatial datasets.”

[ChatSpatial computes RNA velocity]

🤖 ChatSpatial: “RNA velocity computation is complete for your brain tissue. Here is what I found:

RNA Velocity Results:

2,651 spots analyzed
Velocity computed for 18,078 genes
Mode: Stochastic (captures cell state transitions)
Average velocity magnitude: 0.34 (moderate dynamics)
Top dynamic genes: Neurod1, Dlx2, Gsx2, Pax6, Tbr1

Key Insights:

Strong velocity patterns in ventricular zones (high neurogenesis)
Directional flow from stem cell regions toward cortical layers
Active differentiation in intermediate zones
Stable expression in mature cortical areas

Would you like to see a velocity field visualization or explore specific regions?”

🎯 What Just Happened?

ChatSpatial automatically:

✅ Detected spliced/unspliced RNA data layers
✅ Chose appropriate scVelo parameters for spatial data
✅ Computed velocity vectors for all genes and spots
✅ Built transition probability matrices
✅ Identified regions with high cellular dynamics
✅ Highlighted biologically relevant patterns

Step 2: Exploring Velocity Patterns

💬 Diving Deeper into Dynamics

👤 You: “This is interesting! Can you show me where the most dynamic changes are happening in my tissue?”

🤖 ChatSpatial: “Absolutely! Let me create visualizations showing cellular dynamics across your brain tissue.”

[Velocity visualization appears]

🤖 ChatSpatial: “Here is your RNA velocity map! 🧠⚡

What you are seeing:

Arrows show velocity direction (where cells are heading)
Arrow length indicates velocity magnitude (speed of change)
Colors represent velocity confidence (red = high, blue = low)
Background shows your spatial tissue structure

Key Biological Patterns:

Ventricular Zone: Strong outward velocities (cells leaving stem state)
Subventricular Zone: Directional flow toward cortex
Intermediate Zone: Cells transitioning to mature neurons
Cortical Plate: Low velocity (mature, stable cells)

This matches known brain development patterns. Your tissue shows active neurogenesis with cells migrating from germinal zones to their final cortical destinations.”

👤 You: “Can you tell me which specific genes are driving these velocity patterns?”

🤖 ChatSpatial: “Good question! Let me identify the top velocity genes and their biological significance:

Top Velocity-Driving Genes:

Neurogenesis & Differentiation:

Neurod1 (Neural differentiation factor) - High velocity in transition zones
Pax6 (Neural progenitor marker) - Decreasing in differentiating areas
Tbr1 (Cortical neuron marker) - Increasing toward cortical plate

Regional Specification:

Dlx2 (Interneuron specification) - Active in ganglionic eminence
Gsx2 (Progenitor maintenance) - High in ventricular zones
Emx2 (Cortical patterning) - Gradient across cortical areas

Migration & Maturation:

Dcx (Neuronal migration) - High in migratory paths
Map2 (Neuronal maturation) - Increasing in cortical layers
Gfap (Glial differentiation) - Active in astrocyte-rich regions

These genes tell the story of brain development: stem cells (Pax6) differentiating into neurons (Neurod1), migrating to cortex (Dcx), and maturing (Map2)!”

💡 Natural Ways to Explore Velocity

Questions you can ask:

“Where are the most dynamic changes happening?”
“Which genes have the highest velocity?”
“Can you show me velocity patterns in specific regions?”
“How confident is the velocity computation?”
“What biological processes do these patterns represent?”

Follow-up explorations:

“Compare velocity between different conditions”
“Show me genes that are speeding up vs slowing down”
“Which areas have the most directional flow?”

Step 3: Trajectory Inference and Pseudotime

💬 Understanding Developmental Paths

👤 You: “This velocity analysis is informative! Now I want to understand the actual developmental trajectories. Can you trace how cells progress through different states?”

🤖 ChatSpatial: “Good next step! Trajectory analysis will show you the developmental paths that cells follow. Let me infer pseudotime and cellular trajectories using the velocity information we just computed.

For trajectory inference, I can use:

CellRank (recommended with velocity data) - Uses velocity to guide trajectory inference
Palantir - Good general-purpose trajectory method
DPT - Simple diffusion-based approach
VELOVI is NOT a trajectory method - it computes velocity which is then used by these methods

Let me start with CellRank since we have velocity data.”

[ChatSpatial performs trajectory analysis]

🤖 ChatSpatial: “I have reconstructed the developmental trajectories in your brain tissue using CellRank, which combines RNA velocity with spatial information.

Trajectory Analysis Results:

Developmental Timeline:

Pseudotime range: 0.0 (earliest) to 1.0 (most mature)
7 distinct cell states identified along trajectories
3 major developmental branches discovered

Cell State Progression:

Neural stem cells (pseudotime 0.0-0.2) → Ventricular zones
Neural progenitors (pseudotime 0.2-0.4) → Subventricular zones
Migrating neurons (pseudotime 0.4-0.6) → Intermediate zones
Cortical neurons (pseudotime 0.6-0.8) → Cortical layers
Mature neurons (pseudotime 0.8-1.0) → Upper cortical layers

Spatial-Temporal Insights:

Trajectories flow from inner (ventricular) to outer (cortical) regions
Multiple parallel paths suggest different neuronal subtypes
Branch points occur at key developmental decision points
Spatial organization mirrors temporal progression

This captures the inside-out pattern of mammalian cortical development.”

👤 You: “This is very informative! Can you show me these trajectories on the actual tissue?”

🤖 ChatSpatial: “Absolutely! Let me create a spatial trajectory map.”

[Spatial-temporal visualization appears]

🤖 ChatSpatial: “Here is your developmental trajectory map! 🌈🧠

What you are seeing:

Color gradient represents pseudotime (purple = early, yellow = late)
Trajectory paths show developmental flows
Branch points indicate cell fate decisions
Arrows show predicted future cell states

Biological Insights:

The radial pattern matches known neural development
Multiple streams suggest parallel differentiation programs
Branch points occur at anatomically correct locations
Timing matches known developmental stages

Your tissue captures a snapshot of brain development in action.”

Step 4: Detailed Velocity Analysis with VELOVI

💬 Deep Learning Approach to Velocity Computation

👤 You: “This traditional approach is good! I am curious about the newer deep learning methods. Can you try VELOVI for better velocity computation?”

🤖 ChatSpatial: “Good idea! VELOVI uses deep learning to compute RNA velocity more accurately by learning complex velocity patterns. It is particularly good for noisy data and can capture subtler velocity dynamics.

Important: VELOVI is specifically for velocity computation - after computing velocities with VELOVI, we will still use trajectory methods like CellRank or Palantir to infer the actual developmental paths. Let me run VELOVI for velocity computation.”

[ChatSpatial runs VELOVI deep learning model]

🤖 ChatSpatial: “VELOVI velocity computation is complete. Here is how the improved velocities compare to traditional scVelo:

VELOVI Velocity Results:

Technical Performance:

Model training: Converged after 800 epochs
Latent dimensions: 10 (captured main variation)
Velocity uncertainty: Computed (shows confidence)
GPU acceleration: Used for faster training

Improved Velocity Quality:

More refined velocity vectors - smoother, less noisy
Better uncertainty estimates - shows which regions are reliable
Subtle pattern discovery - found additional velocity transitions
Improved gene rankings - more biologically relevant

Key Velocity Improvements from scVelo:

Smoother velocity fields - less jagged vectors
Better handling of noise - more robust to technical artifacts
Uncertainty quantification - tells you where to trust velocity estimates
Novel transition detection - discovered intermediate velocity states

When to Use VELOVI for Velocity:

Complex datasets with multiple cell types
Noisy or challenging data
Need for velocity uncertainty estimates
Publication-quality velocity analyses

Now let me use these improved VELOVI velocities with CellRank for trajectory inference!”

[ChatSpatial runs CellRank with VELOVI velocities]

🤖 ChatSpatial: “I have now computed trajectories using the improved VELOVI velocities. The trajectory inference shows cleaner developmental paths due to the more accurate velocity estimates from VELOVI.”

👤 You: “That is helpful! What about the uncertainty - which regions should I trust most?”

🤖 ChatSpatial: “Good question! The uncertainty information is valuable for interpretation:

High Confidence Regions (trust these results):

Ventricular zones: Clear stem cell dynamics
Cortical plate: Well-defined mature states
Major transition zones: Strong velocity signals

Medium Confidence Regions (interpret carefully):

Boundary areas: Mixed cell populations
Low-cell-density spots: Limited data
Tissue edges: Potential artifacts

Lower Confidence Regions (use caution):

Sparse areas: Few cells for reliable inference
Highly heterogeneous spots: Mixed signals
Technical noise regions: Processing artifacts

Practical Recommendations:

Focus biological interpretation on high-confidence regions
Validate uncertain areas with additional experiments
Use confidence maps to guide follow-up studies
Combine with spatial context for better interpretation

This uncertainty information helps you prioritize which findings to pursue experimentally!”

Step 5: Method Comparison and Selection

💬 Choosing the Right Approach

👤 You: “I have seen scVelo and VELOVI now. How do I know which method is best for my research question?”

🤖 ChatSpatial: “Good question! Let me break down when to use each method:

Method Comparison Guide

Velocity Computation Methods

scVelo 🚀

Best for:

First-time RNA velocity analysis
Quick exploratory studies
Well-behaved, clean datasets
Standard developmental processes
Teaching and demonstrations

Advantages:

Fast computation (minutes)
Well-established and validated
Extensive documentation
Works with most datasets
Good for beginners

Limitations:

Can be noisy with complex data
Limited uncertainty quantification
Basic handling of technical artifacts

VELOVI 🧠

Best for:

Complex, multi-lineage systems
Noisy or challenging datasets
Publication-quality velocity computation
Need uncertainty estimates
Detailed research projects

Advantages:

More accurate velocity estimates
Built-in uncertainty quantification
Better noise handling
Can discover subtle patterns
Deep learning powered

Limitations:

Slower (requires training)
More complex to interpret
Higher computational requirements
Newer (less established)

Trajectory Inference Methods

CellRank 🎯

Best for:

When velocity data is available
Complex transition dynamics
Absorbing states identification
Publication-quality trajectory analysis

Advantages:

Uses velocity to guide trajectories
Identifies terminal states automatically
Handles complex branching patterns
Well-validated method

Limitations:

Requires velocity computation first
More computationally intensive
Complex parameter tuning

Palantir 🌟

Best for:

General-purpose trajectory inference
Multi-branched developmental processes
When velocity data is limited
Robust trajectory computation

Advantages:

Works with or without velocity
Handles complex branching well
Good for exploratory analysis
Stable across different datasets

Limitations:

May not utilize velocity optimally
Requires manual root selection
Less integration with spatial info

DPT 📏

Best for:

Simple trajectory inference
Quick exploratory analysis
Linear or simple branching processes
Independent simple method (explicit user choice)

Advantages:

Fast and simple
Well-established method
Minimal parameter tuning
Good for simple systems

Limitations:

Does not use velocity information
Limited for complex trajectories
Less complex than modern methods

My Recommendations:

Typical Workflow:

Velocity Computation (choose one):
- scVelo (quick standard method, 3 modes available)
- VeloVI (accurate deep learning method)
Trajectory Inference (choose one):
- CellRank (best with velocity data from step 1)
- Palantir (general-purpose, works without velocity)
- DPT (simple pseudotime, works without velocity)
- Note: VELOVI is NOT a trajectory method - it provides velocity for step 1

Specific Recommendations:

For Beginners:

Velocity: Start with scVelo to learn concepts
Trajectory: Use CellRank to utilize velocity information

For Publication:

Velocity: Use VELOVI for accurate velocity computation
Trajectory: Use CellRank for detailed trajectory inference

For Exploration:

Velocity: Try both scVelo and VELOVI and compare
Trajectory: Compare CellRank and Palantir results

For Integration:

Velocity: Use scVelo or VeloVI based on your accuracy needs
Trajectory: Use Palantir for robust inference

Your dataset (brain development) would benefit from:

scVelo first → CellRank - quick velocity + trajectory insights
VELOVI → CellRank - refined velocity + trajectory analysis
Compare both workflows - validate findings”

Step 6: Interpreting Biological Significance

💬 Making Sense of Your Results

👤 You: “I have all these velocity and trajectory results now. How do I interpret them biologically? What do they mean for my research?”

🤖 ChatSpatial: “Good question! Let me help you interpret these results in biological context:

🧬 Biological Interpretation Guide

RNA Velocity Patterns Mean:

High Velocity Regions:

Active cellular transitions happening
Developmental hotspots or response zones
Unstable cell states (transitioning)
Key regulatory events occurring

Low Velocity Regions:

Stable, mature cell states
Homeostatic maintenance modes
Terminal differentiation achieved
Quiescent cell populations

Directional Flow Patterns:

Developmental progressions (stem → mature)
Migration pathways (source → destination)
Response cascades (stimulus → response)
Tissue organization principles

Your Brain Development Results Show:

Biological Processes:

Neurogenesis: Active stem cell differentiation
Migration: Radial movement toward cortex
Specification: Cell fate determination
Maturation: Functional neuron development
Organization: Layer-specific positioning

Spatial Organization:

Inside-out development: Earlier neurons in deeper layers
Radial glia scaffolding: Guiding migration paths
Regionalization: Different areas develop differently
Temporal waves: Coordinated developmental timing

Research Implications:

What Your Data Reveals:

Normal development patterns - your tissue is developing correctly
Timing of processes - when different events occur
Spatial coordination - how location affects development
Key regulatory genes - which genes control transitions

Follow-up Research Questions:

Which signaling pathways regulate these transitions?
How do perturbations (disease/drugs) affect these patterns?
What cell-cell interactions guide development?
Which transcription factors drive state changes?

Experimental Validation:

In situ hybridization - validate top velocity genes
Immunostaining - confirm protein expression patterns
Perturbation studies - test predicted regulators
Single-cell validation - verify computational predictions”

👤 You: “This is helpful! Can you suggest specific experiments to validate these findings?”

🤖 ChatSpatial: “Absolutely! Here are targeted experiments to validate your velocity and trajectory findings:

🔬 Validation Experiment Guide

Immediate Validation (1-2 weeks):

1. In Situ Hybridization

Target: Top 5 velocity genes (Neurod1, Pax6, Tbr1, Dlx2, Dcx)
Prediction: Expression gradients matching trajectory paths
Outcome: Validate spatial expression patterns

2. Immunofluorescence

Markers: Pax6 (stem), Tbr1 (neurons), Map2 (mature)
Prediction: Spatial gradient from ventricular → cortical
Outcome: Confirm protein-level trajectory progression

Extended Validation (1-2 months):

3. Time-course Analysis

Method: Multiple developmental timepoints
Prediction: Earlier timepoints show more ventricular velocity
Outcome: Validate temporal dynamics

4. Perturbation Studies

Target: Key velocity regulators (Pax6, Neurod1)
Method: Knockdown/overexpression
Prediction: Altered trajectory patterns
Outcome: Test functional importance

Complex Validation (3-6 months):

5. Live Imaging

Method: Time-lapse of developing tissue
Prediction: Cell movements match trajectory predictions
Outcome: Direct observation of dynamics

6. Single-cell Validation

Method: scRNA-seq time series
Prediction: Individual cells follow predicted paths
Outcome: Cell-level trajectory validation

Priority Ranking:

In situ + IF (essential, quick)
Perturbation (functional validation)
Time-course (temporal validation)
Live imaging (direct validation)

Would you like detailed protocols for any of these experiments?”

Common Questions and Troubleshooting

💬 “My data does not have spliced/unspliced layers. Can I still do velocity analysis?”

🤖 ChatSpatial: “Yes! You have two options:

Option 1: Trajectory Without Velocity

Focus on trajectory analysis using expression patterns only
Methods like Palantir and DPT work without velocity
Still provides pseudotime and developmental insights
Just ask: ‘Can you analyze trajectories without velocity data?’

Option 2: Expression Dynamics

Analyze temporal patterns in gene expression
Use spatial patterns to infer developmental processes
Less direct but still informative”

💬 “The velocity patterns look noisy. How can I improve them?”

🤖 ChatSpatial: “Noisy velocity is common! Try these approaches:

Parameter Optimization:

Ask: ‘Can you try velocity with different parameters?’
Increase filtering stringency
Adjust neighborhood parameters

Method Switching:

Try VELOVI: ‘Can you run VELOVI for smoother results?’
Use different velocity modes: ‘Try deterministic mode’

Data Quality:

Check: ‘What is the quality of my spliced/unspliced data?’
Consider preprocessing improvements

Spatial Smoothing:

Ask: ‘Can you spatially smooth the velocity results?’
Incorporate spatial information more strongly”

💬 “How do I know if my trajectories are biologically meaningful?”

🤖 ChatSpatial: “Good question! Look for these validation signs:

Biological Consistency:

Trajectories follow known developmental patterns
Gene expression changes match literature
Spatial organization makes anatomical sense
Timing aligns with known processes

Technical Quality:

High velocity confidence scores
Consistent results across methods
Stable across parameter changes
Reproducible in replicate samples

Experimental Support:

Marker genes match predicted states
Perturbation studies confirm predictions
Independent datasets show similar patterns

Ask: ‘How confident should I be in these trajectory results?’”

Parameter Selection Guide

🎯 Choosing the Right Parameters

RNA Velocity Parameters:

Mode Selection:
- "stochastic": Good for most analyses (recommended)
- "deterministic": For clean, consistent data  
- "dynamical": For complex temporal dynamics

Number of PCs:
- 30-50: Standard for most datasets
- Higher: For complex, multi-lineage data
- Lower: For simple systems or small datasets

Trajectory Parameters:

Method Selection:
- "cellrank": Best when velocity is available
- "palantir": Good general-purpose method
- "dpt": Simple diffusion-based method

Spatial Weight:
- 0.3-0.5: Balanced spatial-expression integration
- Higher: More spatial influence  
- Lower: More expression-based

VELOVI Parameters (for velocity computation only):

Training Epochs:
- 500-1000: Standard training for velocity computation
- More: For complex velocity dynamics
- Less: For quick velocity exploration

Latent Dimensions:
- 10-20: Standard for most velocity data
- Higher: For very complex velocity systems
- Lower: For simple velocity dynamics

Note: After VELOVI computes velocities, use trajectory methods:
- CellRank (recommended with VELOVI velocities)
- Palantir (general trajectory inference)
- DPT (simple pseudotime)

Next Steps and Additional Applications

Recommended Analysis Workflow

🔄 Complete Velocity + Trajectory Pipeline

Step 1: Choose Velocity Method (Optional but Recommended)

Quick exploration: “Compute RNA velocity with scVelo” (3 modes available)
High accuracy: “Compute RNA velocity with VeloVI” (deep learning velocity)
Note: Both methods require spliced/unspliced RNA layers in your data

Step 2: Trajectory Inference (Required)

With velocity from Step 1: “Infer trajectories with CellRank using the velocity data”
Without velocity: “Infer trajectories with Palantir” (expression-based)
Simple analysis: “Compute pseudotime with DPT” (diffusion-based)

Important: VELOVI computes velocity (Step 1), not trajectories (Step 2)

Step 3: Integration & Validation

Compare methods: “Compare CellRank vs Palantir trajectories”
Spatial validation: “Show trajectory patterns on tissue”
Biological validation: “Identify marker genes along trajectories”

🚀 Continue Your Analysis

Ready to explore more? Try these additional analyses:

“Can you compare velocity patterns between conditions?”
“Show me how trajectories change with drug treatment”
“Which pathways regulate these developmental transitions?”
“Can you predict future cell states based on current velocity?”
“How do spatial constraints affect trajectory progression?”

🔬 Integration with Other Analyses

Combine trajectory analysis with:

Cell Communication Analysis - How signaling guides trajectories
Cell Type Annotation - Identity along trajectories
Visualization Tutorial - Trajectory visualization plots

📚 Additional Topics

Explore further approaches:

Multi-modal trajectory analysis (RNA + protein + chromatin)
Cross-species trajectory comparison
Disease trajectory analysis (normal vs pathological)
Drug response trajectories (treatment effects over time)

Summary

RNA velocity and trajectory analysis reveal the dynamic story hidden in your spatial transcriptomics data. Through natural conversation with ChatSpatial, you can:

Uncover cellular dynamics without complex programming
Discover developmental pathways through intuitive questions
Compare different methods to find the best approach
Interpret biological significance with expert guidance
Design validation experiments based on computational predictions

Remember: Every trajectory tells a story about cellular journeys through development, disease, or response. ChatSpatial helps you read that story and understand its biological significance.

Key Takeaway: The combination of RNA velocity (showing direction) and trajectory analysis (showing paths) provides insights into how cellular processes unfold in space and time. Your tissue is not just a static snapshot - it is a dynamic system in motion, and these analyses help you see that motion clearly.

Start your trajectory exploration today: “Can you analyze the developmental dynamics in my tissue?”