[Paper Review] Design and Analysis Strategies for Pooling in High Throughput Screening: Application to the Search for a New Anti-Microbial

Byran Smucker, Benjamin Brennan+5 more|arXiv (Cornell University)|Feb 18, 2026

Cell Image Analysis Techniques0 citations

TL;DR

The paper compares pooling designs and statistical analyses for high-throughput screening (HTS) of antimicrobials, introduces a lambda-specific Gauss-Lasso approach, and demonstrates its application to identify true hits while controlling false positives.

ABSTRACT

A major public health issue is the growing resistance of bacteria to antibiotics. An important part of the needed response is the discovery and development of new antimicrobial strategies. These require the screening of potential new drugs, typically accomplished using high-throughput screening (HTS). Traditionally, HTS is performed by examining one compound per well, but a more efficient strategy pools multiple compounds per well. In this work, we study several recently proposed pooling construction methods, as well as a variety of pooled high-throughput screening analysis methods, in order to provide guidance to practitioners on which methods to use. This is done in the context of an application of the methods to the search for new drugs to combat bacterial infection. We discuss both an extensive pilot study as well as a small screening campaign, and highlight both the successes and challenges of the pooling approach.

Motivation & Objective

• Motivate pooling in HTS as a more efficient alternative to one-compound-per-well screening in the search for new antimicrobials.
• Compare three pooling design strategies (CRowS, MAPS, Random) under realistic 384-well formats.
• Evaluate multiple pooled HTS analysis methods, focusing on regularization and thresholding to identify true hits.
• Propose and assess a secondary screening criterion to reduce false positives and sequencing/validation costs.
• Apply pooling designs to a specific antimicrobial target (MtlD inhibition in Salmonella) to illustrate practical implementation and results.

Proposed method

• Model the pooled HTS data as y = beta0 1 + X beta + epsilon with X in {-1,1}^{n x k} indicating compound presence or absence in wells.
• Compare three pool-construction methods: Constrained Row Screening (CRowS), Matrix-Augmented Pooling Strategy (MAPS), and Random Pools, using UE(s^2) and related design criteria.
• Use regularized regression approaches to estimate beta, including Gauss-Lasso (two-stage Lasso followed by OLS refit), non-negative Gauss-Lasso, and elastic net with cross-validation and permutation-based p-value like assessment.
• Introduce lambda-specific thresholding in Gauss-Lasso to leverage known effect directions (inhibitory effects are negative) and improve hit identification.
• Incorporate a secondary screening criterion that requires a compound to inhibit a high proportion of wells containing it, to reduce false positives at the cost of potential power loss.
• Apply these methods to pilot and small-scale HTS campaigns for MtlD inhibition, demonstrating how pooling can identify true hits while controlling false positives.

a Plot of TPR and FPR, as a function of effect size and design size.

Experimental results

Research questions

• RQ1 Which pooling design (CRowS, MAPS, Random) provides the best overall performance under realistic HTS constraints?
• RQ2 Which analysis method (Gauss-Lasso variants, non-negative Gauss-Lasso, Elastic Net) offers the best trade-off between true positive rate and false positive rate for pooled HTS?
• RQ3 Does a lambda-specific thresholding strategy improve hit detection compared with standard Gauss-Lasso or Elastic Net in pooled HTS?
• RQ4 Can a secondary criterion effectively reduce follow-up validation workload while preserving detection of large-effect hits?
• RQ5 How do pooling designs perform in a real antimicrobial screening context targeting MtlD inhibition in Salmonella?

Key findings

• CRowS designs generally perform best or near-best across simulated settings, with advantages in both design criteria (UE(s^2)) and MAPS criteria, and are preferred for flexible, statistically grounded pooling.
• The lambda-specific Gauss-Lasso approach achieves the best overall log(TPR/FPR) performance among analyzed methods, especially with tau_lambda set to max(beta_hat_lambda).
• Elastic Net achieves higher TPR but much higher FPR, indicating a trade-off between discovery and false positives that must be weighed by the researcher.
• Non-negative Gauss-Lasso variants provide competitive performance and may be preferable when effect directions are known to be inhibitory.
• Secondary criteria (e.g., requiring multiple wells per compound to show inhibition) dramatically reduce false positives, at some cost to power, illustrating a practical path to more manageable follow-up validation.
• Proof-of-concept and small screening campaigns show pooling can identify true hits (WT inhibitors not affecting MUT) and distinguish pseud hits, with controlled false-positive rates (~0.3% in spiked experiments).

b Plots of log(TPR/FPR), as a function of effect size and design size.