miniQuant-multi — Multi-Platform Isoform Quantification

miniQuant-multi is a multi-platform RNA isoform quantification tool that integrates sequencing data generated from different platforms or multiple datasets from the same platform, including long-read (LR) and short-read (SR) sequencing. It uses a mixed Bayesian network framework, with parameter estimation performed via the expectation-maximization (EM) algorithm. miniQuant-multi also provides an identifiability analysis function that computes identifiability indicators to quantify how well isoforms can be distinguished from one another.

Table of Contents

• Features
• Installation
• Data Preparation
• Quick Start
• Subcommand: quantify
• Subcommand: cal_identifiability
• Running Modes
• Output Files

---

Features

• Flexible input support: accepts Illumina, Oxford Nanopore, and PacBio SAM files, allowing single-platform, multi-platform, or multi-sample analyses.
• Joint isoform quantification: integrates multiple LR and/or SR datasets using a mixed Bayesian network with EM-based parameter estimation.
• Community-based parallel EM: partitions genes into independent communities and resolves each community separately for efficient inference.
• Flexible sample weighting: supports equal (default), user-defined, or quality-based sample weights estimated from unique mapping rates.
• Isoform identifiability analysis: uses TrEESR.py and identifiability.py to compute identifiability indicators that measure isoform distinguishability from annotation structure and read data.

---

Installation

Create conda environment (recommended)

git clone https://github.com/calm109/miniQuant-multi.git
cd miniQuant-multi
conda env create -f environment.yml
conda activate miniQuant-multi
bash run_miniquant_multi_example.sh

---

Data Preparation

• Install minimap2 (v2.24) and bowtie2 (v2.4.1).

Long-read alignment data mapped to the reference genome in SAM format

• Use minimap2 to map long reads data.

# cDNA / ONT direct RNA
minimap2 -t 3 -ax splice -uf --secondary=no \
    reference_genome.fa long_reads.fastq.gz > LR.sam

# PacBio Iso-Seq
minimap2 -t 3 -ax splice:hq --secondary=no \
    reference_genome.fa long_reads.fastq.gz > LR.sam

Short-read alignment data mapped to the reference transcriptome in SAM format.

• Use Bowtie2 to map short reads data.

# bowtie2 (transcriptome alignment)
bowtie2-build -f transcriptome.fa bowtie2_index

bowtie2 -q --phred33 --sensitive --dpad 0 --gbar 99999999 --mp 1,1 --np 1 --score-min L,0,-0.1 -I 1 -X 1000 --no-mixed --no-discordant -p 10 -k 200 \
-x bowtie2_index -1 reads_R1.fastq.gz -2 reads_R2.fastq.gz > SR.sam

---

Quick Start

BASE_DIR="/path/miniQuant-multi"
MINIQUANT_DIR="$BASE_DIR/isoform_quantification"
DATA_DIR="$BASE_DIR/example"
GTF="$DATA_DIR/annotation.gtf"
OUTPUT_DIR="$BASE_DIR/output"
THREADS=1

# Input files from example_data
SR_SAMS=("$DATA_DIR/SR.sam" "$DATA_DIR/SR.sam")
LR_SAMS=("$DATA_DIR/LR.sam" "$DATA_DIR/LR.sam")

LR_ARGS=""
if [[ ${#LR_SAMS[@]} -gt 0 ]]; then
    LR_ARGS="-lrsam ${LR_SAMS[*]}"
fi
SR_ARGS=""
if [[ ${#SR_SAMS[@]} -gt 0 ]]; then
    SR_ARGS="-srsam ${SR_SAMS[*]}"
fi
mkdir -p "$OUTPUT_DIR"

# 1. Quantify analysis
python "$MINIQUANT_DIR/main.py" quantify \
    -gtf "$GTF" \
    -o "$OUTPUT_DIR" \
    -t "$THREADS" \
    $LR_ARGS \
    $SR_ARGS

# 2. Identifiability analysis
python "$MINIQUANT_DIR/main.py" cal_identifiability \
    -gtf "$GTF" \
    -o "$OUTPUT_DIR" \
    -t "$THREADS" \
    $SR_ARGS \
    $LR_ARGS

---

Subcommand: quantify

python main.py quantify -gtf <GTF> -o <OUTPUT> [options]

Required arguments

Argument	Description
-gtf / --gtf_annotation_path	Path to GTF annotation file
-o / --output_path	Output directory

Input data (at least one required)

Argument	Description
-lrsam / --long_read_sam_path	Path(s) to long-read SAM file(s). Multiple files: -lrsam LR1.sam LR2.sam
-srsam / --short_read_sam_path	Path(s) to short-read SAM file(s). Multiple files: -srsam SR1.sam SR2.sam

Key optional arguments

Argument	Default	Description
-t / --threads	1	Number of threads
--EM_SR_num_iters	200	Number of SR EM iterations
--isoform_start_end_site_tolerance	20	Tolerance (bp) for matching LR start/end sites to isoform boundaries
--junction_site_tolerance	5	Tolerance (bp) for matching splice junction sites

---

Subcommand: cal_identifiability

Computes identifiability metrics (least eigenvalue, condition numbers, standard deviation, and confidence interval) for all genes after running quantification.

python main.py cal_identifiability -gtf <GTF> -o <OUTPUT> [options]

Required arguments

Argument	Description
-gtf / --gtf_annotation_path	Path to GTF annotation file
-o / --output_path	Output directory

Input data (at least one required)

Argument	Description
-lrsam / --long_read_sam_path	Long-read SAM file(s) (used to build data-driven LR A matrix)
-srsam / --short_read_sam_path	Short-read SAM file(s) (used to build data-driven SR A matrix)

Key optional arguments

Argument	Default	Description
-t / --threads	1	Number of threads
--sr_region_selection	read_length	SR region selection mode: read_length (filter by actual read length), real_data (filter by observed reads)
--lr_region_selection	read_length	LR region selection mode: read_length (filter by global median LR read length), real_data (use all observed regions)
--add_full_length_region	all	Whether to keep zero-read-count regions in LR matrix A: all, nonfullrank, none
--normalize_lr_A	True	Column-normalize LR design matrix A
--normalize_sr_A	True	Column-normalize SR design matrix A
--output_matrix_info	False	Output per-gene A matrices and b vectors to matrix_info/

---

Running Modes

Routing logic

The quantify subcommand automatically selects the computation path based on the input data provided:

Condition	Mode	Algorithm
Only SR provided (any number)	SR-only	EM_hybrid_multi
Only LR provided (any number)	LR-only	EM_hybrid_multi
Both LR and SR provided	Hybrid	EM_hybrid_multi

All inputs route through EM_hybrid_multi (community-based parallel EM), where reads and isoforms form a bipartite graph and connected components (gene communities) are solved independently in parallel. The mode (SR-only, LR-only, or hybrid) is determined automatically from which input files are provided — no manual flag needed.

Alpha (LR/SR balance)

• alpha = 1.0: LR-only quantification
• alpha = 0.0: SR-only quantification
• 0 < alpha < 1: hybrid mode, weighting the contribution of LR vs SR
• Default: 0.5 (equal weighting between LR and SR)

---

Output Files

quantify output

File	Description
Isoform_abundance.out	Main result: isoform expression (TPM and read count) for each sample
matrix_info/	Per-gene A matrix and b vector files (when --output_matrix_info True)

Isoform_abundance.out columns:

Column	Description
isoform_id	Isoform transcript ID
gene_id	Gene ID
TPM	Transcripts per million
read_count	Estimated read count
length	Transcript length (bp)

cal_identifiability output

File	Description
kvalues_gene.out	Per-gene A-matrix-based metrics (annotation structure only, no EM required): k-value, regular/generalized condition number, and A matrix dimensions for each SR/LR sample plus Hybrid (when both LR and SR are provided)
kvalues_isoform.out	Per-isoform attributes: isoform length, exon count, and gene membership. Join with kvalues_gene.out on Gene for A-matrix condition number metrics
identifiability.tsv	Per-gene identifiability metrics (requires EM quantification results): minimum positive singular value of the restricted Jacobian (_rJacobi_sigma_min_pos), restricted Jacobian k-value (_rJacobi_k_value), restricted Fisher information matrix metrics including minimum eigenvalue (_rfisher_lambda_min), log10 eigenvalue condition number (_rfisher_lambda_cond), log10 of T-th root of absolute determinant (_rfisher_det = log10(|det(F_tan)|^(1/T)), where T is the number of isoforms; nan when singular), and identifiability flag (_rfisher_identifiable), for each sample and platform (LR, SR, Hybrid)
isoform_SE_CI.tsv	Per-isoform standard errors and 95% confidence intervals of EM-estimated abundance (θ), computed from the restricted Fisher information matrix
SR_singular_values.out / SR_{i}_singular_values.out	Per-gene singular values of the SR A matrix (one file per sample when multiple SR files are provided)
LR_singular_values.out / LR_{i}_singular_values.out	Per-gene singular values of the LR A matrix (one file per sample when multiple LR files are provided)

Genes with empty region sets (e.g., single-exon genes) appear in kvalues.out but are excluded from identifiability.tsv.

---