# Continuous, Low Latency Estimation of the Size and Shape of Single Proteins from Real-Time Nanopore Data

## TL;DR

A new method enables real-time analysis of nanopore data to instantly determine the size and shape of single proteins during experiments.

## Contribution

The novel Two-Sliding Window algorithm allows real-time processing of nanopore data with millisecond response times for protein characterization.

## Key findings

- The TSW algorithm processes data at 40 MB/s on a 1.8 GHz CPU, enabling real-time analysis.
- The TSW algorithm accurately detects events with dwell times greater than 1.5 × (fc)−1.
- Integration of a Naïve Bayes classifier allows real-time classification of protein mixtures for immediate feedback control.

## Abstract

Existing approaches for nanopore sensing typically analyze
resistive
pulses from the translocation of individual proteins through the nanopore
after completing the experiment. This approach foregoes instantaneous
protein identification and precludes real-time experimental control.
Here, we introduce a method for the analysis of real-time nanopore
data capable of characterizing the size and approximate shape of proteins
within a millisecond response time during data acquisition. The implemented
real-time Two-Sliding Window (TSW) peak detection algorithm makes
it possible, for the first time, to filter data, process baselines,
and extract resistive pulse information during nanopore recordings
using a stream of single data points. This approach achieves a computational
throughput of 40 MB/s on a 1.8 GHz laptop CPU. We compared the accuracy
of dwell time determination of the TSW algorithm with an established
offline threshold searching (TS) algorithm, using simulated resistive
pulses. The TSW algorithm accurately extracted events with dwell times
greater than 1.5 times the reciprocal of the system’s cutoff
frequency, 1.5 × (fc)−1. Moreover, we verify
experimentally that integrating the TSW algorithm into the data acquisition
process makes it possible to determine the approximate shape and volume
of proteins within low-millisecond response times. Finally, by integrating
a Naïve Bayes classifier, the system achieves real-time classification
of protein mixtures, allowing for instant feedback control for manipulating
single proteins during or immediately after translocation. This analysis
also improves data storage efficiency during recordings with a high
sampling rate.