Finite-size effects in transcript sequencing count distribution : its power-law correction necessarily precedes downstream normalization and comparative analysis

Background: Though earlier works on modelling transcript abundance from vertebrates to lower eukaroytes have specifically singled out the Zip’s law, the observed distributions often deviate from a single power-law slope. In hindsight, while power-laws of critical phenomena are derived asymptotically...

Full description

Saved in:
Bibliographic Details
Main Authors: Ng, Hong-Kiat, Wong, Wing-Cheong, Tantoso, Erwin, Soong, Richie, Eisenhaber, Frank
Other Authors: School of Computer Science and Engineering
Format: Article
Language:English
Published: 2018
Subjects:
Online Access:https://hdl.handle.net/10356/86009
http://hdl.handle.net/10220/45276
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
id sg-ntu-dr.10356-86009
record_format dspace
institution Nanyang Technological University
building NTU Library
country Singapore
collection DR-NTU
language English
topic Finite-size Effects
Nyquist Sampling Criterion
spellingShingle Finite-size Effects
Nyquist Sampling Criterion
Ng, Hong-Kiat
Wong, Wing-Cheong
Tantoso, Erwin
Soong, Richie
Eisenhaber, Frank
Finite-size effects in transcript sequencing count distribution : its power-law correction necessarily precedes downstream normalization and comparative analysis
description Background: Though earlier works on modelling transcript abundance from vertebrates to lower eukaroytes have specifically singled out the Zip’s law, the observed distributions often deviate from a single power-law slope. In hindsight, while power-laws of critical phenomena are derived asymptotically under the conditions of infinite observations, real world observations are finite where the finite-size effects will set in to force a power-law distribution into an exponential decay and consequently, manifests as a curvature (i.e., varying exponent values) in a log-log plot. If transcript abundance is truly power-law distributed, the varying exponent signifies changing mathematical moments (e.g., mean, variance) and creates heteroskedasticity which compromises statistical rigor in analysis. The impact of this deviation from the asymptotic power-law on sequencing count data has never truly been examined and quantified. Results: The anecdotal description of transcript abundance being almost Zipf’s law-like distributed can be conceptualized as the imperfect mathematical rendition of the Pareto power-law distribution when subjected to the finite-size effects in the real world; This is regardless of the advancement in sequencing technology since sampling is finite in practice. Our conceptualization agrees well with our empirical analysis of two modern day NGS (Next-generation sequencing) datasets: an in-house generated dilution miRNA study of two gastric cancer cell lines (NUGC3 and AGS) and a publicly available spike-in miRNA data; Firstly, the finite-size effects causes the deviations of sequencing count data from Zipf’s law and issues of reproducibility in sequencing experiments. Secondly, it manifests as heteroskedasticity among experimental replicates to bring about statistical woes. Surprisingly, a straightforward power-law correction that restores the distribution distortion to a single exponent value can dramatically reduce data heteroskedasticity to invoke an instant increase in signal-to-noise ratio by 50% and the statistical/detection sensitivity by as high as 30% regardless of the downstream mapping and normalization methods. Most importantly, the power-law correction improves concordance in significant calls among different normalization methods of a data series averagely by 22%. When presented with a higher sequence depth (4 times difference), the improvement in concordance is asymmetrical (32% for the higher sequencing depth instance versus 13% for the lower instance) and demonstrates that the simple power-law correction can increase significant detection with higher sequencing depths. Finally, the correction dramatically enhances the statistical conclusions and eludes the metastasis potential of the NUGC3 cell line against AGS of our dilution analysis. Conclusion: The finite-size effects due to undersampling generally plagues transcript count data with reproducibility issues but can be minimized through a simple power-law correction of the count distribution. This distribution correction has direct implication on the biological interpretation of the study and the rigor of the scientific findings.
author2 School of Computer Science and Engineering
author_facet School of Computer Science and Engineering
Ng, Hong-Kiat
Wong, Wing-Cheong
Tantoso, Erwin
Soong, Richie
Eisenhaber, Frank
format Article
author Ng, Hong-Kiat
Wong, Wing-Cheong
Tantoso, Erwin
Soong, Richie
Eisenhaber, Frank
author_sort Ng, Hong-Kiat
title Finite-size effects in transcript sequencing count distribution : its power-law correction necessarily precedes downstream normalization and comparative analysis
title_short Finite-size effects in transcript sequencing count distribution : its power-law correction necessarily precedes downstream normalization and comparative analysis
title_full Finite-size effects in transcript sequencing count distribution : its power-law correction necessarily precedes downstream normalization and comparative analysis
title_fullStr Finite-size effects in transcript sequencing count distribution : its power-law correction necessarily precedes downstream normalization and comparative analysis
title_full_unstemmed Finite-size effects in transcript sequencing count distribution : its power-law correction necessarily precedes downstream normalization and comparative analysis
title_sort finite-size effects in transcript sequencing count distribution : its power-law correction necessarily precedes downstream normalization and comparative analysis
publishDate 2018
url https://hdl.handle.net/10356/86009
http://hdl.handle.net/10220/45276
_version_ 1681041388172476416
spelling sg-ntu-dr.10356-860092020-03-07T11:48:58Z Finite-size effects in transcript sequencing count distribution : its power-law correction necessarily precedes downstream normalization and comparative analysis Ng, Hong-Kiat Wong, Wing-Cheong Tantoso, Erwin Soong, Richie Eisenhaber, Frank School of Computer Science and Engineering Finite-size Effects Nyquist Sampling Criterion Background: Though earlier works on modelling transcript abundance from vertebrates to lower eukaroytes have specifically singled out the Zip’s law, the observed distributions often deviate from a single power-law slope. In hindsight, while power-laws of critical phenomena are derived asymptotically under the conditions of infinite observations, real world observations are finite where the finite-size effects will set in to force a power-law distribution into an exponential decay and consequently, manifests as a curvature (i.e., varying exponent values) in a log-log plot. If transcript abundance is truly power-law distributed, the varying exponent signifies changing mathematical moments (e.g., mean, variance) and creates heteroskedasticity which compromises statistical rigor in analysis. The impact of this deviation from the asymptotic power-law on sequencing count data has never truly been examined and quantified. Results: The anecdotal description of transcript abundance being almost Zipf’s law-like distributed can be conceptualized as the imperfect mathematical rendition of the Pareto power-law distribution when subjected to the finite-size effects in the real world; This is regardless of the advancement in sequencing technology since sampling is finite in practice. Our conceptualization agrees well with our empirical analysis of two modern day NGS (Next-generation sequencing) datasets: an in-house generated dilution miRNA study of two gastric cancer cell lines (NUGC3 and AGS) and a publicly available spike-in miRNA data; Firstly, the finite-size effects causes the deviations of sequencing count data from Zipf’s law and issues of reproducibility in sequencing experiments. Secondly, it manifests as heteroskedasticity among experimental replicates to bring about statistical woes. Surprisingly, a straightforward power-law correction that restores the distribution distortion to a single exponent value can dramatically reduce data heteroskedasticity to invoke an instant increase in signal-to-noise ratio by 50% and the statistical/detection sensitivity by as high as 30% regardless of the downstream mapping and normalization methods. Most importantly, the power-law correction improves concordance in significant calls among different normalization methods of a data series averagely by 22%. When presented with a higher sequence depth (4 times difference), the improvement in concordance is asymmetrical (32% for the higher sequencing depth instance versus 13% for the lower instance) and demonstrates that the simple power-law correction can increase significant detection with higher sequencing depths. Finally, the correction dramatically enhances the statistical conclusions and eludes the metastasis potential of the NUGC3 cell line against AGS of our dilution analysis. Conclusion: The finite-size effects due to undersampling generally plagues transcript count data with reproducibility issues but can be minimized through a simple power-law correction of the count distribution. This distribution correction has direct implication on the biological interpretation of the study and the rigor of the scientific findings. ASTAR (Agency for Sci., Tech. and Research, S’pore) Published version 2018-07-26T09:15:03Z 2019-12-06T16:14:15Z 2018-07-26T09:15:03Z 2019-12-06T16:14:15Z 2018 Journal Article Wong, W.-C., Ng, H.-K., Tantoso, E., Soong, R., & Eisenhaber, F. (2018). Finite-size effects in transcript sequencing count distribution : its power-law correction necessarily precedes downstream normalization and comparative analysis. Biology Direct, 13(1), 2-. https://hdl.handle.net/10356/86009 http://hdl.handle.net/10220/45276 10.1186/s13062-018-0204-y en Biology Direct © 2018 The Author(s). Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated. 26 p. application/pdf