pith. sign in

arxiv: 1907.11911 · v1 · pith:XALZGHFMnew · submitted 2019-07-27 · 💻 cs.LG · stat.ML

REP: Predicting the Time-Course of Drug Sensitivity

Cheng Qian , Amin Emad , Nicholas D. Sidiropoulos This is my paper

Pith reviewed 2026-05-24 14:51 UTC · model grok-4.3

classification 💻 cs.LG stat.ML
keywords drug response predictiontime-course gene expressionrecursive predictiontensor completionmultiple sclerosisinterferonmachine learning
0
0 comments X

The pith

A recursive framework called REP predicts drug response at every stage of long-term treatment from initial gene expression levels.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces REP, a recursive prediction method that forecasts drug sensitivity over the full course of therapy by feeding prior response estimates back into the model along with time-course gene expression measurements. Earlier work typically used only one or two snapshots of gene data, which cannot capture how drug effects evolve dynamically. REP adds a recursive loop to carry forward previous predictions and applies tensor completion both to clean noisy or incomplete gene expression records and to impute unseen values. Experiments on interferon-treated multiple sclerosis patients show the approach can generate multi-step forecasts from early measurements alone. A reader would care because many therapies require repeated adjustments that depend on how sensitivity changes over months or years rather than at isolated moments.

Core claim

REP employs a built-in recursive structure that exploits the intrinsic time-course nature of the data and integrates past values of drug responses for subsequent predictions. It also incorporates tensor completion that can not only alleviate the impact of noise and missing data, but also predict unseen gene expression levels. These advantages enable REP to estimate drug response at any stage of a given treatment from some GELs measured in the beginning of the treatment.

What carries the argument

Recursive structure that feeds prior drug-response predictions back into the model, combined with tensor completion on the gene-expression tensor.

If this is right

  • Drug response at later treatment stages can be estimated without new measurements after the initial time points.
  • Tensor completion reduces the effect of noise and missing entries while also filling in unseen gene expression values.
  • The method directly supports modeling of long-term therapies where sensitivity changes over many time points.
  • Predictions become possible at arbitrary intermediate stages rather than only before or after treatment.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same recursive-plus-tensor pattern could be tested on time-course data from other chronic conditions that require ongoing drug adjustment.
  • If the recursion remains stable, the framework might be combined with streaming clinical measurements to update forecasts in real time.
  • Extending the tensor completion step to include additional modalities such as protein or metabolite levels could further reduce reliance on frequent sampling.

Load-bearing premise

The recursive integration of past drug response values together with tensor completion on gene expression levels will produce accurate multi-step forecasts without the recursion amplifying noise or the tensor model introducing systematic bias on the specific patient cohort.

What would settle it

On the 53-patient interferon multiple-sclerosis dataset, generate REP forecasts for later time points using only the initial gene expression measurements and compare them to the actual measured drug-response values; systematic divergence at later stages would falsify the claim.

Figures

Figures reproduced from arXiv: 1907.11911 by Amin Emad, Cheng Qian, Nicholas D. Sidiropoulos.

Figure 1
Figure 1. Figure 1: Sketch view of the proposed method. In (a), Step 1) shows the raw data [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Prediction accuracy comparison on raw data, where the percentage [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: ROC curves. 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 Prediction accuracy Percentage of missing [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Prediction accuracy of REP with estimated GEL. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
read the original abstract

The biological processes involved in a drug's mechanisms of action are oftentimes dynamic, complex and difficult to discern. Time-course gene expression data is a rich source of information that can be used to unravel these complex processes, identify biomarkers of drug sensitivity and predict the response to a drug. However, the majority of previous work has not fully utilized this temporal dimension. In these studies, the gene expression data is either considered at one time-point (before the administration of the drug) or two timepoints (before and after the administration of the drug). This is clearly inadequate in modeling dynamic gene-drug interactions, especially for applications such as long-term drug therapy. In this work, we present a novel REcursive Prediction (REP) framework for drug response prediction by taking advantage of time-course gene expression data. Our goal is to predict drug response values at every stage of a long-term treatment, given the expression levels of genes collected in the previous time-points. To this end, REP employs a built-in recursive structure that exploits the intrinsic time-course nature of the data and integrates past values of drug responses for subsequent predictions. It also incorporates tensor completion that can not only alleviate the impact of noise and missing data, but also predict unseen gene expression levels (GELs). These advantages enable REP to estimate drug response at any stage of a given treatment from some GELs measured in the beginning of the treatment. Extensive experiments on a dataset corresponding to 53 multiple sclerosis patients treated with interferon are included to showcase the effectiveness of REP.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces REP, a recursive prediction framework that leverages time-course gene expression data via a built-in recursive structure (integrating past drug response predictions) and tensor completion (to handle noise, missing values, and impute unseen GELs) to forecast drug response at arbitrary future stages of long-term treatment, given only initial gene expression levels. The approach is demonstrated on a cohort of 53 multiple sclerosis patients treated with interferon.

Significance. If the recursive forecasts prove stable without noise amplification and the tensor model yields unbiased imputations on this cohort, the work would offer a practical advance in modeling dynamic gene-drug interactions for personalized long-term therapy prediction, moving beyond single- or two-timepoint analyses common in prior studies.

major comments (2)
  1. [Abstract / Experiments] Abstract and Experiments: The central claim that REP can produce accurate multi-step forecasts at any treatment stage from initial GELs depends on the recursion not amplifying errors and tensor completion avoiding systematic bias; however, no analysis of forecast horizon, error growth rates, ablation of the recursive component, or stability metrics on the 53-patient cohort is described, leaving this load-bearing premise unverified.
  2. [Method] Method description: The recursive integration of predicted drug responses as inputs for subsequent steps is presented as an advantage, but without reported checks for closed-loop stability or comparison to non-recursive baselines, it is unclear whether the framework mitigates or exacerbates prediction drift over multiple time steps.
minor comments (2)
  1. [Abstract] The abstract states 'extensive experiments' but supplies no quantitative metrics, baseline comparisons, or error analysis; these details should be added to the Experiments section for reproducibility.
  2. [Method] Notation for GELs, tensor completion, and the recursive update rule should be formalized with equations to clarify the integration of past responses.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and will revise the manuscript accordingly to strengthen the validation of the recursive framework.

read point-by-point responses

  1. Referee: [Abstract / Experiments] Abstract and Experiments: The central claim that REP can produce accurate multi-step forecasts at any treatment stage from initial GELs depends on the recursion not amplifying errors and tensor completion avoiding systematic bias; however, no analysis of forecast horizon, error growth rates, ablation of the recursive component, or stability metrics on the 53-patient cohort is described, leaving this load-bearing premise unverified.

    Authors: We agree that the current experiments emphasize overall prediction performance on the 53-patient cohort without dedicated ablation or stability analyses. In the revised manuscript we will add an ablation comparing the full recursive REP to a non-recursive variant, along with plots of error accumulation over increasing forecast horizons and basic stability metrics (e.g., variance of successive predictions) to directly address this concern. revision: yes

  2. Referee: [Method] Method description: The recursive integration of predicted drug responses as inputs for subsequent steps is presented as an advantage, but without reported checks for closed-loop stability or comparison to non-recursive baselines, it is unclear whether the framework mitigates or exacerbates prediction drift over multiple time steps.

    Authors: The manuscript motivates the recursive structure by its ability to exploit time-course dependencies, yet we acknowledge the absence of explicit closed-loop stability checks or direct baseline comparisons. We will include these comparisons and stability diagnostics in the revised methods and experiments sections. revision: yes

Circularity Check

0 steps flagged

No circularity; framework claims rest on described architecture without self-referential reductions

full rationale

The abstract presents REP as a novel framework employing a recursive structure to integrate past drug responses and tensor completion to handle noise/missing GELs and impute unseen values, thereby enabling multi-stage predictions from initial measurements. No equations, parameter-fitting steps, or self-citations are supplied that would make any claimed prediction equivalent to its inputs by construction (e.g., no fitted parameter renamed as forecast or ansatz smuggled via prior work). The central claim is an empirical modeling approach whose validity is asserted via experiments on the 53-patient cohort; absent any visible reduction of outputs to inputs, the derivation chain is self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no free parameters, axioms, or invented entities are specified in the provided text.

pith-pipeline@v0.9.0 · 5807 in / 932 out tokens · 17767 ms · 2026-05-24T14:51:30.468893+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

23 extracted references · 23 canonical work pages

  1. [1]

    A community effort to assess and improve drug sensitivity prediction algorithms,

    J. C. Costello, L. M. Heiser, E. Georgii, M. G ¨onen, M. P. Menden, N. J. Wang, M. Bansal, P. Hintsanen, S. A. Khan, J.-P. Mpindi et al. , “A community effort to assess and improve drug sensitivity prediction algorithms,” Nature biotechnology, vol. 32, no. 12, p. 1202, 2014

    work page 2014

  2. [2]

    Predicting cancer drug response using a recommender system,

    C. Suphavilai, D. Bertrand, and N. Nagarajan, “Predicting cancer drug response using a recommender system,” Bioinformatics, vol. 34, no. 22, pp. 3907–3914, 2018

    work page 2018

  3. [3]

    From gene expression to drug response: A collaborative filtering approach,

    C. Qian, N. D. Sidiropoulos, M. Amiridi, and A. Emad, “From gene expression to drug response: A collaborative filtering approach,” in ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2019, pp. 7465–7469

    work page 2019

  4. [4]

    A novel heterogeneous network-based method for drug response prediction in cancer cell lines,

    F. Zhang, M. Wang, J. Xi, J. Yang, and A. Li, “A novel heterogeneous network-based method for drug response prediction in cancer cell lines,” Scientific reports, vol. 8, no. 1, p. 3355, 2018

    work page 2018

  5. [5]

    Cancer cell lines as patient avatars for drug response prediction,

    U. McDermott, “Cancer cell lines as patient avatars for drug response prediction,” Nature genetics, vol. 50, no. 10, p. 1350, 2018

    work page 2018

  6. [6]

    Studying and modelling dynamic biological processes using time-series gene expression data,

    Z. Bar-Joseph, A. Gitter, and I. Simon, “Studying and modelling dynamic biological processes using time-series gene expression data,” Nature Reviews Genetics , vol. 13, no. 8, p. 552, 2012

    work page 2012

  7. [7]

    Elastic net- based prediction of ifn- β treatment response of patients with multiple sclerosis using time series microarray gene expression profiles,

    A. Fukushima, M. Sugimoto, S. Hiwa, and T. Hiroyasu, “Elastic net- based prediction of ifn- β treatment response of patients with multiple sclerosis using time series microarray gene expression profiles,” Scien- tific reports, vol. 9, no. 1, p. 1822, 2019

    work page 2019

  8. [8]

    Clustering of time-course gene expression data using a mixed-effects model with b-splines,

    Y . Luan and H. Li, “Clustering of time-course gene expression data using a mixed-effects model with b-splines,” Bioinformatics, vol. 19, no. 4, pp. 474–482, 2003

    work page 2003

  9. [9]

    Transcription-based prediction of response to ifn β using supervised computational methods,

    S. E. Baranzini, P. Mousavi, J. Rio, S. J. Caillier, A. Stillman, P. Vil- loslada, M. M. Wyatt, M. Comabella, L. D. Greller, R. Somogyi et al., “Transcription-based prediction of response to ifn β using supervised computational methods,” PLoS biology, vol. 3, no. 1, p. e2, 2004. 7

    work page 2004

  10. [10]

    Alignment and classification of time series gene expression in clinical studies,

    T.-h. Lin, N. Kaminski, and Z. Bar-Joseph, “Alignment and classification of time series gene expression in clinical studies,” Bioinformatics, vol. 24, no. 13, pp. i147–i155, 2008

    work page 2008

  11. [11]

    Y .-H. Taguchi, “Identification of candidate drugs using tensor- decomposition-based unsupervised feature extraction in integrated anal- ysis of gene expression between diseases and drugmatrix datasets,” Scientific reports, vol. 7, no. 1, p. 13733, 2017

    work page 2017

  12. [12]

    Non-negative matrix and tensor factorization based classification of clinical microarray gene expression data,

    Y . Li and A. Ngom, “Non-negative matrix and tensor factorization based classification of clinical microarray gene expression data,” in 2010 IEEE international conference on bioinformatics and biomedicine (BIBM) . IEEE, 2010, pp. 438–443

    work page 2010

  13. [13]

    Tricluster: an effective algorithm for mining coherent clusters in 3d microarray data,

    L. Zhao and M. J. Zaki, “Tricluster: an effective algorithm for mining coherent clusters in 3d microarray data,” in Proceedings of the 2005 ACM SIGMOD international conference on Management of data. ACM, 2005, pp. 694–705

    work page 2005

  14. [14]

    Mining coherent gene clusters from gene-sample-time microarray data,

    D. Jiang, J. Pei, M. Ramanathan, C. Tang, and A. Zhang, “Mining coherent gene clusters from gene-sample-time microarray data,” in Proceedings of the tenth ACM SIGKDD international conference on Knowledge discovery and data mining . ACM, 2004, pp. 430–439

    work page 2004

  15. [15]

    Time series cluster kernel for learning similarities between multivariate time series with missing data,

    K. Ø. Mikalsen, F. M. Bianchi, C. Soguero-Ruiz, and R. Jenssen, “Time series cluster kernel for learning similarities between multivariate time series with missing data,” Pattern Recognition, vol. 76, pp. 569–581, 2018

    work page 2018

  16. [16]

    A review on missing value imputation algorithms for microarray gene expression data,

    K. Moorthy, M. Saberi Mohamad, and S. Deris, “A review on missing value imputation algorithms for microarray gene expression data,” Current Bioinformatics, vol. 9, no. 1, pp. 18–22, 2014

    work page 2014

  17. [17]

    Comparison of discrimination methods for the classification of tumors using gene expression data,

    S. Dudoit, J. Fridlyand, and T. P. Speed, “Comparison of discrimination methods for the classification of tumors using gene expression data,” Journal of the American statistical association , vol. 97, no. 457, pp. 77–87, 2002

    work page 2002

  18. [18]

    Missing value imputation methods for gene-sample-time microarray data analysis,

    Y . Li, A. Ngom, and L. Rueda, “Missing value imputation methods for gene-sample-time microarray data analysis,” in 2010 IEEE Symposium on Computational Intelligence in Bioinformatics and Computational Biology. IEEE, 2010, pp. 1–7

    work page 2010

  19. [19]

    Gene expression prediction using low-rank matrix completion,

    A. Kapur, K. Marwah, and G. Alterovitz, “Gene expression prediction using low-rank matrix completion,” BMC bioinformatics, vol. 17, no. 1, p. 243, 2016

    work page 2016

  20. [20]

    Tensor decomposition for signal processing and machine learning,

    N. D. Sidiropoulos, L. De Lathauwer, X. Fu, K. Huang, E. E. Papalex- akis, and C. Faloutsos, “Tensor decomposition for signal processing and machine learning,” IEEE Transactions on Signal Processing , vol. 65, no. 13, pp. 3551–3582, July 2017

    work page 2017

  21. [21]

    A block coordinate descent method for regularized multiconvex optimization with applications to nonnegative tensor fac- torization and completion,

    Y . Xu and W. Yin, “A block coordinate descent method for regularized multiconvex optimization with applications to nonnegative tensor fac- torization and completion,” SIAM Journal on imaging sciences , vol. 6, no. 3, pp. 1758–1789, 2013

    work page 2013

  22. [22]

    Pharmacokinetic/pharmacodynamic parameters: rationale for antibacterial dosing of mice and men,

    W. A. Craig, “Pharmacokinetic/pharmacodynamic parameters: rationale for antibacterial dosing of mice and men,” Clinical infectious diseases , vol. 26, no. 1, pp. 1–10, 1998

    work page 1998

  23. [23]

    k-nearest neighbor models for microarray gene expression analysis and clinical outcome prediction,

    R. Parry, W. Jones, T. Stokes, J. Phan, R. Moffitt, H. Fang, L. Shi, A. Oberthuer, M. Fischer, W. Tonget al., “k-nearest neighbor models for microarray gene expression analysis and clinical outcome prediction,” The pharmacogenomics journal , vol. 10, no. 4, p. 292, 2010

    work page 2010