FEAST

Simulation and interpolation of spatial transcriptomics from parameter cloud

FEAST: Simulation and interpolation of spatial transcriptomics from parameter cloud

Status: Active project, ready for RECOMB 2026 conference submission

Role: First author, Independent Research Project

Institution: Prof. Maizie Zhou Lab, BME & CS Department, Vanderbilt University

Duration: August 2024 – Present (Remote and on-site collaboration)

Preprint PDF

Project Overview

FEAST (FEAture-space based modeling for Spatial Transcriptomics) is a computational infrastructure that models ST data within a parameter cloud — a latent manifold encoding gene-level mean, variance, and sparsity. By sampling and perturbing this manifold, FEAST generates high-fidelity synthetic slices with tunable spatial and transcriptional variation, enabling systematic evaluation of clustering, deconvolution, and spatial alignment algorithms.

Beyond two dimensions, FEAST performs 3D parameter-cloud interpolation guided by optimal transport and benchmark alignment, reconstructing continuous tissue architectures while preserving molecular coherence. Together, these capabilities establish FEAST as a foundational platform for standardized benchmarking, data augmentation, and 3D reconstruction in spatial transcriptomics.

Code: GitHub · PyPI

Technical Innovation

Statistical Modeling:

  • Parameter Cloud Representation: Models gene-level mean, variance, and sparsity in a unified latent manifold
  • C-vine Copula: Captures nonlinear and asymmetric dependencies among gene parameters
  • Flexible Count Models: Supports Poisson, Negative Binomial, ZIP, and ZINB distributions

Simulation Framework:

  • Single Slice Simulation: Generates high-fidelity 2D ST slices with controllable alterations
  • Expression Alteration: Tunable perturbations in mean, variance, and sparsity for robustness testing
  • Spatial Transformation: Controlled geometric transformations for alignment benchmarking

3D Interpolation:

  • Optimal Transport: Wasserstein barycenter-based interpolation between parameter clouds
  • Alignment-guided Coordinates: Transport-guided spatial coordinate interpolation
  • Volumetric Reconstruction: Generates intermediate slices for continuous 3D tissue architectures

Key Results

  • High-fidelity simulation with near-perfect correlation for gene means and variances across multiple ST platforms (10x Visium, MERFISH, OpenST, Slide-seqV2, Xenium, Stereo-seq)
  • Clustering benchmarking revealed algorithmic sensitivities under controlled expression and sparsity perturbations
  • Alignment evaluation demonstrated Spateo’s robustness over SPACEL under geometric transformations
  • 3D interpolation achieved >0.9 correlation with ground-truth experimental slices in leave-one-out validation

Complete Research Cycle

This project represents a comprehensive research experience including:

  • Literature Review: Extensive survey of spatial transcriptomics simulation methods
  • Idea Exploration: Creative hypothesis generation and initial concept development
  • Failure Analysis: Learning from initial approaches that didn’t meet expectations
  • Method Validation: Rigorous testing and validation of novel approaches
  • Model Construction: Implementation of robust, scalable computational framework
  • Benchmark Development: Creating comprehensive evaluation metrics and comparison studies
  • Application Studies: Demonstrating utility across diverse biological scenarios
  • Manuscript Preparation: Scientific writing and presentation of results

Professional Development

The project has provided extensive training in:

  • Public Presentation: Regular presentation of progress at different project milestones
  • Scientific Communication: Development of clear, compelling research narratives
  • Independent Research: Self-directed project management and problem-solving
  • International Collaboration: Remote and on-site work with US-based research team

Expected Impact

FEAST provides the research community with:

  • Standardized Benchmarking: Controllable ground-truth datasets for evaluating clustering, deconvolution, and alignment algorithms
  • Data Augmentation: Generate realistic synthetic ST data for method development and validation
  • 3D Reconstruction: Interpolate missing slices to reconstruct continuous tissue volumes from sparse experimental sections
  • Platform-agnostic Design: Works across diverse ST technologies with unified parameter-cloud representation

Conference Submission

The project is currently being prepared for submission to RECOMB 2026, one of the premier conferences in computational biology, demonstrating the high quality and significance of this research.

This project showcases independent research capability and innovative thinking in computational biology.

References