Combining machine learning with high-content imaging to infer ciprofloxacin susceptibility in clinical isolates of Salmonella Typhimurium

Aug 3, 2026

Upload by Ivy

Machine learning is combined with high-content imaging to infer ciprofloxacin susceptibility in clinical isolates of Salmonella Typhimurium. The material describes time- and concentration-dependent changes in bacterial morphology and length under ciprofloxacin exposure, using multiple fluorescence stains to capture cellular phenotypes. A random forest approach is evaluated through out-of-bag error analysis and feature importance ranking. Classifier performance is further benchmarked across training, validation, and test sets using accuracy, sensitivity, specificity, precision, F1 score, and AUC.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages8

Reading Time20 min

AccessFree to download

Supplementary Figures

Bacterial length and morphology under ciprofloxacin exposure

Random forest differentiation of resistant and susceptible isolates

Isolates used for classifier training, validation, and testing

Supplementary Tables

Evaluation of multiple machine learning methods

Isolates and whole-genome sequencing accession numbers

Combining machine learning with high-content imaging to infer ciprofloxacin susceptibility in clinical isolates of Salmonella Typhimurium | | | | | | | | | --- | --- | --- | --- | --- | --- | --- | a b D23580 | Bacteria Length (µm) | 10
8
6
4
2
10
8
6
4
2 | D23580 SL1344
SL1344gyrA VNS20081

5 10 15 20 25 5 10 15 20 25 | | --- | --- | --- | | Time [hour] | | | 2 6 10 14 18 22 24 Time [hour] c V NS20081 | |
6 |
10 |
14 |
18 |
22 |
24 | | --- | --- | --- | --- | --- | --- | --- | Concentration 0x 1x 2x 4x Concentration 0x 1x 2x 4x Time [hour] Supplementary Fig. 2. Change in bacterial length over time and ciprofloxacin exposure. (a) Dynamics of bacterial length (vertical axis) over 24h of ciprofloxacin exposure (horizontal axis) . (b) and (c) Morphologies of S. Typhimurium D23580 and VNS20081, respectively, at different ciprofloxacin relative concentration and exposure time. Relative concentrations (x MIC) increase from top to bottom and exposure time increases from left to right. Red, green, and blue fluorescence are from CSA, SYTOX Green, and DAPI stains, respectively. a | Error Rate | 0.08
0.06
0.04
0.02
0.00 |
Number of Trees | | --- | --- | --- | b Mean DAPI Intensity SG Profile 2.2 CSA Profile 2.2 SG Radial Relative Deviation Susceptible SG Intensity Mean Resistant SG Threshold Compactness 50 SG Symmetry 15 CSA Profile 1.2 SG Profile 1.2 DAPI Intensity StdDev Supplementary Fig. 4. Random forest model differentiation of resistant and susceptible isolates at 0xMIC-22h. (a) OOB error rates of random forest models trained from data at 0xMIC-22h. Horizontal axis is number of decision trees used in random forest ensembles, and vertical axis is respective OOB error rate. (b) Spider chart for the ten most important features to differentiate susceptible (yellow) and resistant (dark purple) isolates. Data was transferred to z-score, and each line is a datapoint. D23580 VNS20081 SL1344gyrA SL1344 13 other isolates Supplementary Fig. 5. Isolates evaluated in machine learning classifier testing, training, and validation sets. Venn diagram of isolates used to train machine learning classifiers. Green circle represents the four main isolates and orange circle represents 16 isolates to test the generalization of the classifiers. Supplementary Table 1. Evaluation of different machine learning methods to identify resistant S. Typhimurium on training, validation, and test sets. Training set Validation set Test set | Method | Accura
cy | Sensiti
vity | Specifi
city | Precisi
on | F1
score AUC | | Accura
cy | Sensiti
vity | Specifi
city | Precisi
on | F1
score AUC | | Accura
cy | Sensiti
vity | Specifi
city | Precisi
on | F1
score AUC | | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | Naïve
Bayes
KNN
SVM Random forest
CatBoost
Neural
network | 0.79±0
.06
0.86±0
.04
0.87±0
.04
0.71±0
.06
0.64±0
.06
0.83±0
.06 | 0.64±0
.12
0.79±0
.06
0.84±0
.07
0.99±0
.02
1.00±0
.01
0.83±0
.12 | 0.93±0
.04
0.92±0
.04
0.91±0
.05
0.45±0
.11
0.31±0
.11
0.83±0
.10 | 0.89±0
.07
0.90±0
.05
0.89±0
.06
0.62±0
.06
0.57±0
.06
0.82±0
.08 | 0.74±0
.10
0.84±0
.05
0.86±0
.05
0.76±0
.04
0.73±0
.04
0.82±0
.07 | 0.90±0
.05
0.91±0
.05
0.92±0
.05
0.92±0
.04
0.87±0
.05
0.91±0
.04 | 0.73±0
.13
0.80±0
.11
0.80±0
.09
0.73±0
.10
0.68±0
.09
0.88±0
.08 | 0.61±0
.25
0.73±0
.20
0.77±0
.15
0.99±0
.03
1.00±0
.02
0.88±0
.11 | 0.86±0
.14
0.87±0
.11
0.84±0
.12
0.45±0
.19
0.32±0
.17
0.8

FAQs

How is ciprofloxacin susceptibility assessed using high-content imaging in the study?

The study tracks bacterial morphology and length over 24 hours under varying ciprofloxacin relative concentrations, quantified using fluorescence stains. These imaging-derived features are then used as inputs to machine learning classifiers.

What does the random forest evaluation show for resistant versus susceptible isolates?

It reports out-of-bag (OOB) error rates across different numbers of decision trees and provides a spider chart of the ten most important discriminative features separating susceptible and resistant isolates.

Which machine learning methods are compared, and how is performance measured?

Multiple methods are compared, including Naïve Bayes, KNN, SVM, random forest, CatBoost, and neural networks. Performance is reported on training, validation, and test sets using metrics such as accuracy, sensitivity, specificity, precision, F1 score, and AUC.

Aug 3, 2026

Upload by Ivy