NanoBoost

A Wavelet Transform-Enhanced Machine Learning Algorithm for Next-Generation Nanopore Multiplexing

NanoBoost is a machine learning pipeline combining XGBoost with a novel event-specific Discrete Wavelet Transform (DWT) for nanopore translocation event classification. It achieved 91% accuracy in nanoparticle size classification (5 nm vs 10 nm nanospheres) and 99% in shape classification (nanospheres vs nanorods).

Installation

git clone https://github.com/joehart2001/nanoboost.git
cd nanoboost
pip install -e .

For unsupervised learning and plotting dependencies:

pip install -e ".[full]"

Usage

End-to-end run

Edit config.yaml to point to your ABF files and set labels, then:

nanoboost run config.yaml

Step-by-step

# 1. Isolate events from a raw ABF recording
nanoboost preprocess recording.abf events.pkl

# 2. Apply per-event DWT and extract features  (label: 5 or 10 for nm, NR or NS for shape)
nanoboost transform events.pkl features.pkl 10

# 3. Train a classifier across one or more feature sets
nanoboost train features_5nm.pkl features_10nm.pkl --output model.pkl

Use nanoboost <command> --help for full options on any command.

Configuration

config.yaml controls all parameters. Paper-optimal defaults are pre-set:

data:
  files:
    - path: recordings/5nm.abf
      label: 5
    - path: recordings/10nm.abf
      label: 10
  output_dir: results/

preprocessing:
  resistive: false      # true for biphasic events with a resistive (trough) component

transform:
  wavelet: bior3.3      # paper-optimal mother wavelet
  threshold: 0.2        # paper-optimal DWT coefficient threshold

training:
  model: xgboost        # xgboost | rf | svm | dt
  search: random        # random | bayes | sobol | grid

CLI flags override config values for individual commands:

nanoboost transform events.pkl features.pkl 10 --wavelet haar --threshold 0.3
nanoboost train features.pkl --model rf --search bayes

Citation

If you use this code or methodology in your research, please cite this repository (paper details to be added upon publication). Here is a suggested BibTeX entry:

@software{nanoboost,
  author = {Hart, Joseph},
  title = {NanoBoost: A Wavelet Transform-Enhanced Machine Learning Algorithm for Next-Generation Nanopore Multiplexing},
  year = {2025},
  url = {https://github.com/joehart2001/nanoboost}
}

For the complete citation information, see CITATION.cff.

License

MIT — see LICENSE.