eFISHent 0.0.9

RNA FISH oligos/probes design tool.

eFISHent

A command-line based tool to facilitate the creation of eFISHent single-molecule RNA fluorescence in-situ hybridization (RNA smFISH) oligonucleotide probes.

Contents

• Description
• Installation
• Getting Genomes, Annotations, Count Tables
• Workflow
• Usage
• Output
• Full Examples
• FAQ

Description

eFISHent is a tool to facilitate the creation of eFISHent RNA smFISH oligonucleotide probes. Some of the key features of eFISHent are:

• One-command installation — no sudo, Docker, or conda required
• Automatic gene sequence download from NCBI when providing a gene and species name (or pass a FASTA file)
• Parameter presets for common FISH protocols (--preset smfish, merfish, dna-fish, etc.)
• Filtering steps to remove low-quality probes including off-targets, frequently occurring short-mers, secondary structures, etc.
• Mathematical or greedy optimization to ensure highest coverage

Installation

eFISHent is tested on macOS and Linux with Python 3.9+. Works on shared HPC/cluster servers via SSH — no sudo, Docker, or conda needed. For Windows users, we recommend WSL.

Quick Install (Recommended)

A single command installs eFISHent and all dependencies:

curl -LsSf https://raw.githubusercontent.com/BBQuercus/eFISHent/main/install.sh | sh

This will:

1. Download pre-compiled binaries for bowtie2, bowtie, jellyfish, BLAST+, GLPK, and Entrez Direct
2. Install Python and the eFISHent package in an isolated environment
3. Create an efishent command that works without any activation

After installation, restart your shell (or source ~/.zshrc) and you're ready:

efishent --check    # Verify all dependencies
efishent --help     # Show usage

Custom Installation Path

curl -LsSf https://raw.githubusercontent.com/BBQuercus/eFISHent/main/install.sh | sh -s -- --prefix /path/to/install

Updating

efishent --update

This updates both the Python package and verifies all external dependencies are still present and compatible.

Checking Dependencies

efishent --check

Shows the status of all required and optional dependencies with version info.

Using Conda

Alternatively, use conda to manage dependencies:

conda env create bbquercus/efishent
conda activate efishent
pip install efishent

Development Installation

git clone https://github.com/BBQuercus/eFISHent.git
cd eFISHent/

# Install dependencies
./install.sh --deps-only

# Install development version
uv venv && source .venv/bin/activate
uv pip install -e .

Uninstalling

curl -LsSf https://raw.githubusercontent.com/BBQuercus/eFISHent/main/install.sh | sh -s -- --uninstall

Or simply remove the install directory: rm -rf ~/.local/efishent

Getting Genomes, Annotations, Count Tables

For some of the steps described below, you'll need to provide a few key resources unique to your model organism/cell line. Below is a quick-start for human cells (the most common use case), followed by general instructions for other organisms.

Human Quick-Start (HeLa, HEK293, U2OS, etc.)

All human cell lines use the same reference genome and annotation — cell-line-specific differences don't affect probe design at the annotation level.

1. Reference genome (required)

Download the GRCh38 primary assembly from Ensembl:

# Primary assembly excludes haplotype patches that inflate off-target counts
wget https://ftp.ensembl.org/pub/current_fasta/homo_sapiens/dna/Homo_sapiens.GRCh38.dna.primary_assembly.fa.gz
gunzip Homo_sapiens.GRCh38.dna.primary_assembly.fa.gz

2. GTF annotation (required for --intergenic-off-targets, --filter-rrna, expression filtering)

Download from Ensembl:

wget https://ftp.ensembl.org/pub/current_gtf/homo_sapiens/Homo_sapiens.GRCh38.115.gtf.gz
gunzip Homo_sapiens.GRCh38.115.gtf.gz

Ensembl GTFs use gene_biotype while GENCODE uses gene_type — eFISHent supports both. Either source works, but Ensembl is recommended for consistency with the genome FASTA.

3. Reference transcriptome (optional, for BLAST cross-validation and comprehensive rRNA filtering)

Generate from your genome + GTF using gffread, then append rRNA sequences for comprehensive filtering:

# Generate transcriptome from genome + annotation
gffread Homo_sapiens.GRCh38.115.gtf -g Homo_sapiens.GRCh38.dna.primary_assembly.fa -w transcriptome.fa

# Download rRNA sequences (major 18S/28S/5.8S are NOT in standard GTFs)
# 45S pre-rRNA contains 18S, 5.8S, and 28S
efetch -db nucleotide -id NR_003286.4 -format fasta >> transcriptome.fa
# 5S rRNA
efetch -db nucleotide -id NR_023363.1 -format fasta >> transcriptome.fa

Why append rRNA? The major rRNA genes (18S, 28S, 5.8S) exist in ~300 tandem copies in nucleolar organizer regions that are not assembled in GRCh38. They are absent from standard GTFs, so --filter-rrna alone won't catch them. Including them in the transcriptome FASTA ensures the BLAST filter removes probes binding these highly abundant sequences.

4. Count table (optional, for expression-weighted off-target filtering)

Download a HeLa (or your cell line) RNA-seq dataset:

1. Go to GEO and search for your cell line + RNA-seq (e.g., HeLa RNA-seq)
2. Under Supplementary file, download files ending in FPKM, TPM, or RPKMs.txt.gz — not raw counts
3. Alternatively, use Expression Atlas — ensure the file has Ensembl IDs in column 1 and normalized counts in column 2

Putting it all together — recommended command for human smFISH:

# Build indices once
eFISHent \
    --reference-genome Homo_sapiens.GRCh38.dna.primary_assembly.fa \
    --build-indices True

# Design probes with all recommended filters
eFISHent \
    --reference-genome Homo_sapiens.GRCh38.dna.primary_assembly.fa \
    --reference-annotation Homo_sapiens.GRCh38.115.gtf \
    --reference-transcriptome transcriptome.fa \
    --gene-name "ACTB" \
    --organism-name "homo sapiens" \
    --preset smfish \
    --mask-repeats True \
    --intergenic-off-targets True \
    --filter-rrna True \
    --threads 8

Other Organisms

For non-human organisms, follow the same pattern using your organism's Ensembl page or UCSC genome browser:

1. Go to Ensembl or UCSC genome browser
2. Download the genome FASTA (.fa.gz) — prefer primary_assembly if available, otherwise toplevel
3. Download the GTF annotation (.gtf.gz) from the same source
4. Unzip both files with gunzip

Workflow

eFISHent works by iteratively selecting probes passing various filtering steps as outlined below:

1. A list of all candidate probes is generated from an input FASTA file containing the gene sequence. This sequence file can be passed manually or downloaded automatically from NCBI when providing a gene and species name.
2. The first round of filtering removes any probes not passing basic sequence-specific criteria including melting temperature (adjusted for formamide and salt concentrations), GC content, G-quadruplets, homopolymer runs (e.g., AAAAA), and optionally low-complexity regions (dinucleotide repeats, low Shannon entropy).
3. Probes are aligned to the reference genome using Bowtie2 (default, with OligoMiner/Tigerfish parameters for sensitive local alignment) or Bowtie (legacy). Candidates with off-targets are removed. Several optional filters can refine off-target counting:
   • Repeat masking (--mask-repeats): ignores off-target hits in repetitive/low-complexity regions identified by dustmasker
   • Intergenic filtering (--intergenic-off-targets): ignores off-target hits outside annotated genes (requires --reference-annotation)
   • Thermodynamic scoring (--off-target-min-tm): rescues probes whose off-target hits have predicted Tm below the hybridization temperature
   • Expression weighting: removes probes hitting highly expressed genes (requires count table)
4. If a reference transcriptome is provided, probes are BLASTed against expressed transcripts (using TrueProbes parameters) to catch off-target binding that genome alignment alone may miss — e.g., hits to processed mRNAs or splice junctions.
5. The targets are divided into short k-mers and discarded if they appear above a determined threshold in the reference genome using Jellyfish.
6. The secondary structure of each candidate is predicted using a nearest neighbor thermodynamic model and filtered if the free energy is too low (too stable), which could result in motifs hindering hybridization.
7. This gives the set of all viable candidates which are still overlapping. The final step is to use mathematical or greedy optimization to maximize probe non-overlapping coverage across the gene sequence. A gap-filling pass then attempts to place additional probes in any remaining uncovered regions.

flowchart TD
    A["🧬 Gene Sequence<br/><i>FASTA file or NCBI download</i>"] --> B["Generate Candidate Probes<br/><i>Sliding window (min/max length)</i>"]

    B --> C["Basic Filtering<br/><i>TM, GC, homopolymers, low-complexity</i>"]

    C --> D["Genome Alignment<br/><i>Bowtie2 (default) or Bowtie</i><br/><i>+ repeat masking, intergenic, Tm scoring</i>"]

    D --> E{"Transcriptome<br/>provided?"}
    E -- Yes --> F["Transcriptome BLAST<br/><i>Off-target detection (TrueProbes params)</i>"]
    E -- No --> G["K-mer Filtering<br/><i>Jellyfish frequency count</i>"]
    F --> G

    G --> H["Secondary Structure<br/><i>ΔG prediction (RNAstructure)</i>"]

    H --> I["Coverage Optimization<br/><i>Greedy or optimal (MILP) + gap filling</i>"]

    I --> J["🎯 Final Probe Set"]

    style A fill:#e1f5fe
    style J fill:#e8f5e9

Usage

Quick Start

eFISHent --reference-genome <reference-genome> --gene-name <gene> --organism-name <organism>

Index Building

While there is only one main workflow, the slightly more time-intensive index creation step can be run ahead of time. Indexes are unique to each reference genome and can be created using:

eFISHent \
    --reference-genome <path to genome fasta file> \
    --build-indices True

Passing Gene Sequence

The actual probe-creating workflow will then not only require the reference genome but also the sequence against which probes should be designed. Probes can be passed in one of three ways:

• --sequence-file - Path to a fasta file containing the gene sequence
• --ensembl-id (& --organism-name) - Ensembl ID of gene of interest. Will be downloaded from Entrez. The organism name can also be passed to avoid some wonky organism genes that have similar names but isn't required.
• --gene-name & --organism-name - Instead of ensembl ID, both gene and organism name can be provided. The sequence will also be downloaded from Entrez.

Optimization Options

There are two ways in which the final set of probe candidates that passed filtering can be assigned/selected using --optimization-method:

• greedy - Uses the next best possibility in line starting with the first probe. This has a time complexity of O(n) with n being the number of candidates. Therefore, even with very loosely set parameters and a lot of candidates, this will still be very fast. This is the default option.
• optimal - Uses a mathematical optimization model to yield highest coverage (number of nucleotides bound to a gene). This has a time complexity of O(n**2) meaning the more probes there are the exponentially slower it will get. Despite breaking the problem into chunks, this might be restrictively slow. However, you can set a time limit (--optimization-time-limit) to stop the optimization process after a given amount of seconds. The resultant probes will be the best ones found so far.

Off-Target Handling

eFISHent provides multiple layers of off-target detection. These can be combined for maximum specificity:

Genome alignment (default) - Probes are aligned to the reference genome using Bowtie2 (sensitive local alignment with OligoMiner/Tigerfish parameters) or Bowtie (legacy). Use --max-off-targets to set the maximum allowed genome hits (default: 0). Use --aligner bowtie to fall back to the legacy aligner.

Transcriptome BLAST (optional) - Probes are BLASTed against a reference transcriptome to catch off-target binding to expressed transcripts that genome alignment alone may miss (e.g., hits spanning splice junctions). BLAST parameters are derived from TrueProbes (Neuert Lab). Enable by providing:

• --reference-transcriptome - A FASTA file of expressed transcripts (generate from genome + GTF using gffread: gffread annotation.gtf -g genome.fa -w transcriptome.fa)
• --max-transcriptome-off-targets - Maximum transcriptome hits per probe (default: 0)
• --blast-identity-threshold - Minimum % identity for a BLAST hit to count (default: 75)

Repeat masking (optional) - Ignores off-target hits landing in repetitive/low-complexity regions (Alu, LINE, SINE elements, etc.). Uses dustmasker from BLAST+ to identify repeats. This is often the biggest win for rescuing probes — 30–50% of spurious off-target hits can be in repeats:

• --mask-repeats - Enable repeat-masked off-target filtering (default: no)

Intergenic off-target filtering (optional) - Ignores off-target hits that don't overlap any annotated gene. For RNA FISH, intergenic regions don't produce transcripts the probe could bind:

• --intergenic-off-targets - Enable intergenic filtering (default: no, requires --reference-annotation)

Thermodynamic off-target scoring (optional) - Rescues probes whose off-target hits are thermodynamically unstable. Computes predicted Tm for each off-target alignment using the nearest-neighbor model; hits with Tm below the threshold are ignored:

• --off-target-min-tm - Minimum Tm (°C) for an off-target to count. Recommended: set to your hybridization temperature (e.g., 37). Set to 0 to disable (default: 0)

Ribosomal RNA filtering (optional) - Removes probes with off-target hits on ribosomal RNA genes. rRNA constitutes ~80% of cellular RNA, so even weak off-target binding causes intense background signal. When a GTF annotation is provided, eFISHent checks gene_biotype/gene_type for rRNA and Mt_rRNA:

• --filter-rrna - Enable rRNA off-target filtering (default: no, requires --reference-annotation)

Note: Standard GTF files (Ensembl/GENCODE) include 5S and mitochondrial rRNA genes but miss the major 18S/28S/5.8S rRNA — these are in unassembled tandem repeat regions. For comprehensive rRNA filtering, also use --reference-transcriptome with rRNA sequences included. You can download the human 45S pre-rRNA (GenBank NR_003286.4) and 5S rRNA (NR_023363.1) from NCBI, concatenate them with your transcriptome FASTA, and the transcriptome BLAST filter will catch probes binding these sequences.

Expression-weighted filtering (optional) - If genome off-targets are unavoidable, you can weight them by expression level to remove probes hitting highly expressed genes:

• --reference-annotation - A GTF genome annotation file to know which genes correspond to which genomic loci
• --encode-count-table - A csv or tsv file with any normalized RNA-seq count table format (FPKM, TPM, etc.) as well as the Ensembl ID matching the entries in the GTF file
• --max-expression-percentage - The percentage of genes to be excluded sorted based on expression level (using the provided count table)

If you don't have your own RNA-seq dataset, you can download available datasets (make sure you're not using raw, but only normalized counts!) at Gene Expression Omnibus. Search for RNA-seq and the name of your organism/cell line.

General Filtering Parameters

There are a bunch of parameters that can be set to adjust filtering steps:

Parameter	Description
--min-length, --max-length	Probe lengths in nucleotides
--spacing	Minimum distances between probes
--min-tm, --max-tm	Minimum and maximum possible melting temperature (affected by length, GC content, and formamide/Na concentration)
--min-gc, --max-gc	GC content in percentage
--formamide-concentration	Percentage of formamide in buffer
--na-concentration	Sodium ion concentration in mM
--max-homopolymer-length	Maximum homopolymer run length (e.g., AAAAA). Default: 5 (matches OligoMiner). Set to 0 to disable
--filter-low-complexity	Filter probes with dinucleotide repeats or low Shannon entropy regions
--aligner	Alignment tool: bowtie2 (default, OligoMiner/Tigerfish params) or bowtie (legacy)
--kmer-length, --max-kmers	Jellyfish-based short-mer filtering. If candidate probes contain kmers of length --kmer-length that are found more than --max-kmers in the reference genome, the candidate will get discarded
--max-deltag	Predicted secondary structure threshold
--sequence-similarity	Will remove probes that might potentially bind to each other (set the similarity to the highest allowed binding percentage). Will reduce the number of probes because (due to otherwise exceptionally high runtimes) has to be run after optimization. Recommended: 75-85 for stringent designs
--off-target-min-tm	Minimum predicted Tm (°C) for an off-target to count as significant. Set to hybridization temperature to rescue probes with thermodynamically unstable off-targets. Default: 0 (disabled)
--mask-repeats	Ignore off-target hits in repetitive/low-complexity regions (uses dustmasker from BLAST+)
--intergenic-off-targets	Ignore off-target hits outside annotated genes (requires --reference-annotation)
--filter-rrna	Remove probes with off-target hits on rRNA genes (requires --reference-annotation)

Remaining Options

There are a few additional options:

Parameter	Description
--is-plus-strand	Set true/false depending on gene of interest
--is-endogenous	Set true/false depending on gene of interest
--threads	Wherever multiprocessing is available, spawn that many threads. Set this to as many cores as you have available
--save-intermediates	Save all intermediary files. Can be used to gauge which filtering steps are set too aggressively
--verbose	Set to get some more information on progress

Probe Set Analysis

After generating a probe set, you can analyze it in more detail using the --analyze-probeset option. This creates a PDF report with visualizations of various probe characteristics:

eFISHent \
    --reference-genome <path to genome fasta file> \
    --sequence-file <path to gene fasta file> \
    --analyze-probeset <path to probe set fasta file>

The analysis includes:

Plot	Description
Lengths	Distribution of probe lengths
Melting temperatures	Boxplot of calculated Tm values
GC Content	Boxplot of GC percentages
G quadruplet	Count of G-quadruplet motifs per probe
K-mer count	Maximum k-mer frequency in genome
Free energy	Predicted secondary structure stability (ΔG)
Off target count	Number of off-target binding sites per probe
Binding affinity	Probe-to-probe similarity matrix (potential cross-hybridization)
Gene coverage	Visual map of probe positions along the target sequence

The output is saved as <probeset_name>_analysis.pdf in the current directory.

Output

By default eFISHent will output three unique files:

• GENE_HASH.fasta - All probes in FASTA format for subsequent usage
• GENE_HASH.csv - A table containing all probes as well as basic parameters (such as melting temperature)
• GENE_HASH.txt - A configuration file to check which parameters were used during the run as well as the command to start it

GENE is a reinterpreted gene name dependent on the options passed but should be immediately clear as to where it's from. The HASH is a unique set of characters that identifies the parameters passed for the run. This way, if and only if the same parameters are passed again eFISHent doesn't have to rerun anything. All intermediary files during the run will be saved in the same format but will get deleted at the end unless --save-intermediates is set to true.

Full Examples

First, the indexes for the respective genome have to be built:

eFISHent \
    --reference-genome ./hg-38.fa \
    --build-indices True

An example to create 45 to 50-mers for a gene of interest downloaded from Entrez:

eFISHent \
    --reference-genome ./hg-38.fa \
    --gene-name "norad" \
    --organism-name "homo sapiens" \
    --is-plus-strand True \
    --optimization-method optimal \
    --min-length 45 \
    --max-length 50 \
    --formamide-concentration 45 \
    --threads 8

Another example using a custom sequence:

eFISHent \
    --reference-genome ./dm6.fa \
    --sequence-file "./renilla.fasta" \
    --is-endogenous False \
    --threads 8

An example with off-target weighting:

eFISHent \
    --reference-genome ./hg-38.fa \
    --reference-annotation ./hg-38.gtf \
    --ensembl-id ENSG00000128272 \
    --organism-name "homo sapiens" \
    --is-plus-strand False \
    --max-off-targets 5 \
    --encode-count-table ./count_table.tsv \
    --max-expression-percentage 20 \
    --threads 8

An example using advanced off-target filtering for maximum probe yield:

eFISHent \
    --reference-genome ./hg-38.fa \
    --reference-annotation ./hg-38.gtf \
    --sequence-file ./my_gene.fasta \
    --mask-repeats True \
    --intergenic-off-targets True \
    --off-target-min-tm 37 \
    --threads 8

Lastly, an example to analyze an existing probe set:

eFISHent \
    --reference-genome ./hg-38.fa \
    --sequence-file ./my_gene.fasta \
    --analyze-probeset ./my_gene_probes.fasta

FAQ

Have questions? Open an issue on GitHub.