The reviewed record of science sign in
Pith

arxiv: 2606.21092 · v1 · pith:QGP6VO57 · submitted 2026-06-19 · cs.LG · math.OC

BASIL: Bayesian Application for Scientific Iteration and Learning

Kelvin P. Idanwekhai , Valeriia Kaneva , Stefano Menegatti , Alexander Tropsha This is my paper

Reviewed by Pith2026-06-26 14:49 UTCgrok-4.3pith:QGP6VO57open to challenge →

classification cs.LG math.OC
keywords Bayesian optimizationsurrogate modelsacquisition functionsprocess optimizationdesktop applicationmulti-objective optimizationexperimental design
0
0 comments X

The pith

BASIL is a desktop application that uses Bayesian surrogate models and acquisition functions to optimize arbitrary user-defined experimental processes.

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

The paper presents BASIL as a graphical tool that lets users enter experimental inputs, objectives, and existing data to build surrogate models. These models pair with acquisition functions to suggest next steps for single-objective or multi-objective optimization. The application supplies ready-made model templates so that the same interface works across different processes. If the approach holds, researchers could run guided optimization loops without writing custom code for each new problem.

Core claim

BASIL is a desktop application that takes user-supplied input variables, optimization objectives, and legacy data, constructs surrogate models from predefined templates, and applies acquisition functions to direct the next experiments toward the stated goals for both single and multi-objective cases.

What carries the argument

The BASIL graphical interface, which couples predefined surrogate model templates with acquisition functions to propose experimental steps from user inputs and data.

If this is right

  • Any process with defined inputs and measurable outputs can be optimized through the same interface.
  • Both single-objective and multi-objective problems are handled by the supplied acquisition functions.
  • Legacy data can be loaded directly to initialize the surrogate models.
  • Users avoid custom model development by selecting from the provided templates.

Where Pith is reading between the lines

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

  • The tool could shorten iteration cycles in laboratory settings where each trial is costly.
  • Broader adoption would depend on how often the templates match real experimental noise and constraints.
  • Future extensions might link the interface directly to automated lab hardware for closed-loop operation.

Load-bearing premise

That the built-in surrogate model templates and acquisition functions will succeed on any arbitrary process without users needing to create or validate new models.

What would settle it

A concrete experiment in which BASIL's suggestions fail to improve the objective on a process whose input-output relationship lies outside the range covered by its predefined templates.

Figures

Figures reproduced from arXiv: 2606.21092 by Alexander Tropsha, Kelvin P. Idanwekhai, Stefano Menegatti, Valeriia Kaneva.

Figure 1
Figure 1. Figure 1: Campaign creation and management interface. (A) Campaign configuration panel, in which the user defines [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Campaign results visualization interface. The Visualization tab allows visualization of relationships between [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Campaign results explanation interface. The Explanations tab provides functionality to visualize SHAP [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
read the original abstract

We introduce BASIL, a user-friendly desktop application for process optimization. BASIL employs a Bayesian approach, incorporating special acquisition functions that can be used to solve both single and multi-objective optimization problems. It provides a graphical interface that enables users to input their experimental parameters, optimization objectives, and legacy data. This is then used to build surrogate models, which are coupled with acquisition functions to guide and optimize a process towards a desired objective. To facilitate model building, BASIL provides a variety of predefined surrogate model templates. BASIL can be used to optimize any arbitrary experiment or process with known, user-defined input variables, optimization objectives, and defined output.

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 / 1 minor

Summary. The manuscript introduces BASIL, a desktop application for Bayesian optimization of experiments and processes. It provides a GUI for users to specify input variables, optimization objectives, and legacy data, which are used to build surrogate models from a variety of predefined templates. These are coupled with acquisition functions to guide single- and multi-objective optimization, with the central claim that the tool applies to any arbitrary experiment or process with known, user-defined inputs, objectives, and outputs.

Significance. If the predefined surrogate templates and acquisition functions prove effective and appropriate across diverse user-defined problems, BASIL could meaningfully lower barriers to applying Bayesian optimization in experimental sciences by offering an accessible interface without requiring custom model development. This would be a practical contribution to iterative scientific workflows.

major comments (2)
  1. [Abstract] Abstract: The claim that BASIL 'can be used to optimize any arbitrary experiment or process with known, user-defined input variables, optimization objectives, and defined output' is load-bearing for the contribution but rests on the untested assumption that the 'variety of predefined surrogate model templates' will be suitable without custom development; no description of the templates, their assumptions (e.g., stationarity, noise handling, variable types), or limitations is provided.
  2. [Abstract] Abstract: No validation experiments, benchmarks, performance metrics, or error analysis are included to demonstrate effectiveness for arbitrary problems, leaving the central claim of broad applicability unsubstantiated; this is especially pertinent given known sensitivities of Bayesian optimization to problem structure such as dimensionality and discreteness.
minor comments (1)
  1. [Abstract] Abstract: The phrase 'special acquisition functions' is imprecise; naming the specific functions or their distinguishing properties would aid reader understanding.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thoughtful review and constructive comments on the manuscript. We address each major comment below and will revise the manuscript to provide greater clarity on the surrogate templates and to better substantiate the tool's applicability.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The claim that BASIL 'can be used to optimize any arbitrary experiment or process with known, user-defined input variables, optimization objectives, and defined output' is load-bearing for the contribution but rests on the untested assumption that the 'variety of predefined surrogate model templates' will be suitable without custom development; no description of the templates, their assumptions (e.g., stationarity, noise handling, variable types), or limitations is provided.

    Authors: We agree that the manuscript would benefit from an explicit description of the surrogate model templates. In the revised version we will add a dedicated subsection detailing the available templates (primarily Gaussian process variants with standard kernels), their assumptions on stationarity, noise modeling, and variable types (continuous, integer, categorical), as well as known limitations such as sensitivity to high dimensionality. This will allow readers to assess the scope of the broad-applicability claim. revision: yes

  2. Referee: [Abstract] Abstract: No validation experiments, benchmarks, performance metrics, or error analysis are included to demonstrate effectiveness for arbitrary problems, leaving the central claim of broad applicability unsubstantiated; this is especially pertinent given known sensitivities of Bayesian optimization to problem structure such as dimensionality and discreteness.

    Authors: The manuscript presents BASIL primarily as an accessible software tool rather than a new algorithmic contribution. We acknowledge the absence of dedicated validation experiments or benchmarks in the current version. In revision we will incorporate illustrative case studies on standard benchmark functions, report basic performance metrics, and add a discussion of BO sensitivities (dimensionality, discreteness) together with user guidance on when custom surrogate development may be preferable. Full error analysis across arbitrary problems is beyond the scope of a tool-description paper but will be addressed via the added examples. revision: yes

Circularity Check

0 steps flagged

No circularity: paper describes software tool with no derivations or fitted predictions

full rationale

The manuscript introduces BASIL as a desktop application for Bayesian optimization. It states that the tool 'provides a variety of predefined surrogate model templates' and 'can be used to optimize any arbitrary experiment or process with known, user-defined input variables, optimization objectives, and defined output.' No equations, first-principles derivations, parameter-fitting steps, or predictions are presented. The central claim is a capability statement about the software rather than a mathematical result derived from inputs. No self-citation chains, ansatzes, or renamings of known results appear in the provided text. The work is self-contained as a tool description; any concerns about template generality fall under correctness or validation rather than circularity in a derivation chain.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no technical details on models, parameters, or assumptions are provided.

pith-pipeline@v0.9.1-grok · 5644 in / 882 out tokens · 16166 ms · 2026-06-26T14:49:37.521422+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

18 extracted references · 15 canonical work pages · 2 internal anchors

  1. [1]

    Nature Synthesis , volume =

    The Rise of Self-Driving Labs in Chemical and Materials Sciences , author =. Nature Synthesis , volume =. doi:10.1038/s44160-022-00231-0 , urldate =

    work page doi:10.1038/s44160-022-00231-0

  2. [2]

    Steph Ango , urldate =

    File over App , author =. Steph Ango , urldate =

  3. [3]

    and Daulton, Samuel and Letham, Benjamin and Wilson, Andrew Gordon and Bakshy, Eytan , year = 2020, month = dec, number =

    Balandat, Maximilian and Karrer, Brian and Jiang, Daniel R. and Daulton, Samuel and Letham, Benjamin and Wilson, Andrew Gordon and Bakshy, Eytan , year = 2020, month = dec, number =. doi:10.48550/arXiv.1910.06403 , urldate =. arXiv , keywords =:1910.06403 , primaryclass =

    work page doi:10.48550/arxiv.1910.06403 2020

  4. [4]

    Bennett, J. A. and Orouji, N. and Khan, M. and Sadeghi, S. and Rodgers, J. and Abolhasani, M. , year = 2024, month = mar, journal =. Autonomous Reaction. doi:10.1038/s44286-024-00033-5 , urldate =

    work page doi:10.1038/s44286-024-00033-5 2024

  5. [5]

    Journal of Chemical Information and Modeling , volume =

    Bertelsen, S. Journal of Chemical Information and Modeling , volume =. doi:10.1021/acs.jcim.4c02240 , urldate =

    work page doi:10.1021/acs.jcim.4c02240

  6. [6]

    Journal of Machine Learning Research , volume =

    D. Journal of Machine Learning Research , volume =

  7. [7]

    Digital Discovery , volume =

    Fitzner, Martin and. Digital Discovery , volume =. doi:10.1039/D5DD00050E , urldate =

    work page doi:10.1039/d5dd00050e

  8. [8]

    A Tutorial on Bayesian Optimization

    Frazier, Peter I. , year = 2018, month = jul, number =. A. doi:10.48550/arXiv.1807.02811 , urldate =. arXiv , keywords =:1807.02811 , primaryclass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1807.02811 2018

  9. [9]

    Bayesian Optimization for Adaptive Experimental Design: A Review

    Greenhill, Stewart and Rana, Santu and Gupta, Sunil and Vellanki, Pratibha and Venkatesh, Svetha , year = 2020, journal =. Bayesian. doi:10.1109/ACCESS.2020.2966228 , urldate =

    work page doi:10.1109/access.2020.2966228 2020

  10. [10]

    Nature Communications , volume =

    Towards a Fully Automated Algorithm Driven Platform for Biosystems Design , author =. Nature Communications , volume =. doi:10.1038/s41467-019-13189-z , urldate =

    work page doi:10.1038/s41467-019-13189-z

  11. [11]

    and Shastry, Shriarjun and Hurst, Morgan R

    Idanwekhai, Kelvin P. and Shastry, Shriarjun and Hurst, Morgan R. and Minzoni, Arianna and Barbieri, Eduardo and Remmler, Luke and Muratov, Eugene N. and Daniele, Michael A. and Menegatti, Stefano and Tropsha, Alexander , year = 2026, journal =. Adaptive. doi:10.1002/bit.70159 , urldate =

    work page doi:10.1002/bit.70159 2026

  12. [12]

    eLife , volume =

    Robotic Search for Optimal Cell Culture in Regenerative Medicine , author =. eLife , volume =. doi:10.7554/eLife.77007 , urldate =

    work page doi:10.7554/elife.77007

  13. [13]

    Lundberg, Scott M and Lee, Su-In , year = 2017, volume =. A. Advances in

    2017

  14. [14]

    Surrogate modeling for Bayesian optimization beyond a single Gaussian process

    Lu, Qin and Polyzos, Konstantinos D. and Li, Bingcong and Giannakis, Georgios B. , year = 2022, publisher =. Surrogate Modeling for. doi:10.48550/ARXIV.2205.14090 , urldate =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2205.14090 2022

  15. [15]

    and Parlane, Fraser G

    MacLeod, Benjamin P. and Parlane, Fraser G. L. and Rupnow, Connor C. and Dettelbach, Kevan E. and Elliott, Michael S. and Morrissey, Thomas D. and Haley, Ted H. and Proskurin, Oleksii and Rooney, Michael B. and Taherimakhsousi, Nina and Dvorak, David J. and Chiu, Hsi N. and Waizenegger, Christopher E. B. and Ocean, Karry and Mokhtari, Mehrdad and Berlingu...

    work page doi:10.1038/s41467-022-28580-6 2022

  16. [16]

    doi:10.26434/chemrxiv.5953606.v1 , urldate =

    Roch, Lo. doi:10.26434/chemrxiv.5953606.v1 , urldate =

    work page doi:10.26434/chemrxiv.5953606.v1

  17. [17]

    Digital Discovery , volume =

    A Self-Driving Laboratory Designed to Accelerate the Discovery of Adhesive Materials , author =. Digital Discovery , volume =. doi:10.1039/D2DD00029F , urldate =

    work page doi:10.1039/d2dd00029f

  18. [18]

    Bayesian reaction optimization as a tool for chemical synthesis

    Bayesian Reaction Optimization as a Tool for Chemical Synthesis , author =. Nature , volume =. doi:10.1038/s41586-021-03213-y , urldate =

    work page doi:10.1038/s41586-021-03213-y