Virus Detection in Multiplexed Nanowire Arrays using Hidden Semi-Markov models

Shalini Ghosh, Patrick Lincoln, Christian Petersen, Alfonso ValdesarXiv 2014 · OpenReview

Abstract

In this paper, we address the problem of real-time detection of viruses docking to nanowires, especially when multiple viruses dock to the same nano-wire. The task becomes more complicated when there is an array of nanowires coated with different antibodies, where different viruses can dock to each coated nanowire at different binding strengths. We model the array response to a viral agent as a pattern of conductance change over nanowires with known modifier --- this representation permits analysis of the output of such an array via belief network (Bayes) methods, as well as novel generative models like the Hidden Semi-Markov Model (HSMM).

Introduction

Nanowire field-effect transistors (FETs) offer a promising path for label-free, real-time virus detection. In these systems, nanowires are coated with specific antibodies; when a virus binds to an antibody, it changes the wire's electrical conductance. However, practical implementation faces three primary hurdles: cross-reactivity between similar viral families, significant signal noise (including baseline drift and common-mode interference), and the complexity of interpreting multiplexed array data. This paper proposes a robust framework using Bayesian networks and Hidden Semi-Markov Models (HSMMs) to overcome these challenges.

Methods

The authors propose a multi-layered approach to signal processing and inference. To handle non-linear baseline drifts, they implement a histogram detrending method. By identifying the most frequent conductance value (the "mode") in a sliding window, they can estimate the local mean more robustly than a standard moving average, which is often biased by the binding events themselves.

Histogram-based detrending method used to identify local conductance modes and robustly estimate the baseline.

For multiplexed arrays, the system is modeled as a Bayesian network, allowing for the calculation of the posterior probability of a specific virus being present based on the collective pattern of conductance changes. The core algorithmic contribution is the use of Hidden Semi-Markov Models (HSMM). Unlike standard HMMs, HSMMs allow for arbitrary duration modeling. By using a multiphase Coxian distribution to model the duration of a virus docking event, the system can distinguish between legitimate binding events and short-lived noise spikes.

Key Results

The HSMM successfully identified docking events even when trained on incomplete or noisy labels, effectively "filling in the gaps" where simple thresholding failed. A critical insight was the exploitation of correlated noise; because wires in an array share the same environment, they often share common-mode noise. By cross-correlating signals, the authors demonstrated that they could extract and subtract this noise, significantly improving the signal-to-noise ratio (SNR).

Signal processing results showing the extraction of binding events by correlating signals across the nanowire array to subtract common-mode noise.

In simulations, the Bayesian approach correctly identified viral mixtures and flagged "unknown" viruses by recognizing atypical binding patterns. The histogram-based detrending also proved superior to global curve fitting for removing persistent shifts caused by sequential binding events.

Conclusion

While the models perform well on simulations and limited experimental sets, scaling these probabilistic models to massively multiplexed arrays will require further computational optimization. Future work must also focus on establishing rigorous calibration phases to ensure low false-alarm rates in complex clinical samples. Despite these challenges, the integration of duration-based modeling via HSMMs represents a significant step forward in reliable, real-time pathogen identification.

Summary updated

Abstract

In this paper, we address the problem of real-time detection of viruses docking to nanowires, especially when multiple viruses dock to the same nano-wire. The task becomes more complicated when there is an array of nanowires coated with different antibodies, where different viruses can dock to each coated nanowire at different binding strengths. We model the array response to a viral agent as a pattern of conductance change over nanowires with known modifier --- this representation permits analysis of the output of such an array via belief network (Bayes) methods, as well as novel generative models like the Hidden Semi-Markov Model (HSMM).
BibTeX
@misc{ghosh2014virus,
  author = {Shalini Ghosh and Patrick Lincoln and Christian Petersen and Alfonso Valdes},
  title = {{Virus Detection in Multiplexed Nanowire Arrays using Hidden Semi-Markov models}},
  year = {2014},
  eprint = {1407.4490},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1407.4490},
}
Figure