Description

HyLoRD (Hybrid Long Read Deconvolution) is a cell type deconvolution tool. It utilitises a hybrid (rather than reference-based/free) approach to deconvolving ONT long read sequencing data.

Introduction

Epigenetic cell type deconvolution is widely used to dissect cellular heterogeneity in bulk tissue samples, with applications in EWAS and biomarker discovery. Traditional methods rely on microarray or short-read bisulfite sequencing (WGBS) data, but long-read sequencing platforms (particularly ONT) offer distinct advantages, including:

Native epigenetic detection: Direct calling of 5mC and 5hmC via basecallers (e.g., Guppy, Dorado, Remora), reducing experimental noise.
Improved genomic coverage: Long reads span more CpG sites, potentially improving deconvolution accuracy.

Despite these advances, no existing deconvolution tool is optimized for long-read methylation data. HyLoRD fills this gap by combining reference-based and reference-free approaches, enabling robust proportion estimation even for cell types absent in the user provided reference methylation matrix.

Methodological Approach

HyLoRD formulates deconvolution as a quadratic programming problem (QPP) optimizing cell proportions under biological constraints (e.g., non-negativity, sum-to-one). Key features include:

Hybrid Deconvolution
- Incorporates prior knowledge from reference methylomes while allowing inference of novel cell types.
- Mitigates biases from incomplete reference datasets.
Computational Efficiency
- Utilizes the Goldfarb-Idnani solver
Multi-Signature Support
- Optional integration of 5hmC signals, improving resolution in cell types with differential hydroxymethylation (particularly useful when deconvolving bulk data from brain tissue for example)

Advantages Over Existing Methods

Accuracy: Leverages long-read methylation calls, which are less susceptible to PCR bias and coverage dropouts.

Flexibility: Compatible with sparse reference datasets, critical given the limited availability of cell-sorted ONT data.

Interpretability: Quadratic programming provides transparent, deterministic solutions.

Applications

HyLoRD is particularly suited for:

Validating cell sorting efficiency in single-cell experiments.
Deconvolving neuronal tissues, where 5hmC signals provide discriminative power.
Analysing archival or low-input samples where long-read sequencing is advantageous.

Installation

Prebuilt binary

Download the latest release from the releases page and run:

# Extract binary

tar -xzvf hylord-[version]-[distribution].tar.gz

Install from source

The build process for HyLoRD is carried out via CMake, however the process is made easier via a Makefile wrapper. To install:

make CMAKE_BUILD_TYPE=Release

sudo make install

To install to the non-default location

# Note that the path provided MUST NOT be a relative path, but a full path
make CMAKE_BUILD_TYPE=Release CMAKE_INSTALL_PREFIX=path/to/hylord
sudo make install

After completing this process make sure that path/to/hylord is on your $PATH with which hylord. If not found, you will need to update your PATH environment variable (details of which can be found here).

You can verify that the installation completed by running:

hylord --version

See BUILD.md for further details, including dependencies.

Running

To get started with HyLoRD, you will need the required input files.

After this you can run HyLoRD with:

hylord -r path/to/reference_matrix.bed path/to/bedmethyl.bed

Further details

Generally when using HyLoRD, the command would look something like:

hylord \
  -r path/to/reference_matrix.bed \
  -l path/to/cell_list.txt \
  -c path/to/cpg_list.bed \
  --additional-cell-types 3 \
  path/to/bedmethyl.bed

Further command line options are explained by running:

hylord -h

Documentation

Full documentation for HyLoRD can be found here. You can also head over to BUILD.md for details on how to build these pages locally with doxygen.

Table of Contents

Table of contents