A proof-of-concept implementation of BL-FlowSOM

BL-FlowSOM is an improved version of FlowSOM that uses batch learning, which is available on Sony’s Spectral Flow Analysis (SFA) - Life sciences Cloud Platform . This repository contains a proof-of-concept implementation of BL-FlowSOM algorithm to reproduce clustering result of it. Please note that the clustering result of BL-FlowSOM and this proof-of-concept implementation writen in Python are almost identical, although there are some minor differences that are assumed to be due to in processing systems of the programming languages.

Overview

This proof-of-concept implementation takes csv file as input, and outputs the trained SOM vector and the classified result of input events to SOM as clustering result. Each row of the input file is one event, and each column is each parameter data of the input dataset.

Requirements

To run this project, you need:

• Python 3.11.0 or later
• pip 25.0 or later

Setup

1. Clone this repository:

   git clone https://github.com/sony/bl-flowsom_poc

2. Move to the cloned directory:

   cd bl-flowsom_poc/

3. Install the required libraries:

   pip install -r requirements.txt

Data preparation

Input data file must be in csv format without headers and can be generated as follows, for example, using fcs file as input and R flowCore package:

write.table(exprs(read.FCS("FP7000_34c.fcs", truncate_max_range = FALSE))[,1:34], "data/FP7000_34c.csv", sep=",", row.names=F, col.names=F)
write.table(exprs(read.FCS("Levine_32dim.fcs", truncate_max_range = FALSE))[,5:36], "data/Levine_32dim.csv", sep=",", row.names=F, col.names=F)
write.table(exprs(read.FCS("Levine_13dim.fcs", truncate_max_range = FALSE))[,1:13], "data/Levine_13dim.csv", sep=",", row.names=F, col.names=F)
write.table(exprs(read.FCS("Samusik_01.fcs", truncate_max_range = FALSE))[,9:47], "data/Samusik_01.csv", sep=",", row.names=F, col.names=F)
write.table(exprs(read.FCS("Samusik_all.fcs", truncate_max_range = FALSE))[,9:47], "data/Samusik_all.csv", sep=",", row.names=F, col.names=F)

For more information on the fcs files listed above, see Evaluation of FlowSOM and BL-FlowSOM clustering result.

Usage

Perform the proof-of-concept code as follows:

python ./build_batchSOM.py -d 10 -s 1 -i ./data/FP7000_34c.csv -o results/FP7000_34c_clust.csv -c results/FP7000_34c_codes.csv

Arguments for build_batchSOM.py are as follows:

• -h: show help
• -d: witdth and height of the SOM grid
• -s: seed [optional]
• -i: events data file in csv format (input file)
• -o: classified result of input events to SOM in csv format (output file)
• -c: trained SOM vector in csv format (output file)

License

This repository is licensed under the Creative Commons BY-NC-SA 4.0 License.