FITs

FITs reconstructs true signals in highly sparse and noisy single-cell open chromatin read-count matrices using an ensemble of imputation trees to improve analyses of regulatory regions.

Key Features:

• Two-Phase Imputation Process: Constructs multiple imputation trees in two phases, with each tree producing one or two imputed versions of the original read-count matrix and supporting parallel execution across processors.
• Noise Reduction and Signal Recovery: Mitigates high dropout rates and stochastic noise in single-cell open chromatin profiles by leveraging multiple imputation trees to reduce bias in restoration.
• Enhanced Accuracy for Cell-Type-Specific Activity: Preserves signals at genomic sites exhibiting cell-type-specific activity during imputation.
• Generalized Applicability to Sparse Matrices: Applicable to highly sparse read-count matrices derived from single-cell open-chromatin profiles from both in vitro and in vivo samples.
• Utility in Minority Cell Analysis: Recovers signals from minority or rare cell types within heterogeneous populations.

Scientific Applications:

• Visualization and Classification: Enables visualization and classification of open chromatin regions to assess cellular heterogeneity.
• Enhancer Detection: Improves accuracy of enhancer detection from single-cell open chromatin data.
• Pathway Enrichment Scoring: Enhances calculation of pathway enrichment scores to identify active biological pathways.
• Chromatin Interaction Prediction: Aids prediction of chromatin interactions and inference of three-dimensional genome organization.

Methodology:

Constructs multiple imputation trees in a two-phase process where each tree generates one or two imputed read-count matrices; tree construction can be executed in parallel across processors and multiple trees are used to mitigate bias in signal restoration.