Frontiers in microbiology · 2026
Population dynamics analysis of an industrial methanotrophic consortium based on Methylococcus capsulatus KN2 using AI technologies.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Zakhartsev Maksim V, Pavlov Dmitriy A, Oshkin Igor Y, Saltykova Victoria A, Kochkarov Azret A
Аннотация
BACKGROUND/INTRODUCTION: Monitoring the population dynamics of industrial methanotrophic bacterial consortia is critical for optimization of single-cell protein (SCP) production from natural gas. Traditional manual microscopy is labor-intensive, subjective, and limited in throughput. AI-based computer vision offers a promising alternative for automated, quantitative analysis of cell morphotypes in mixed cultures.
METHODS: A convolutional neural network (YOLO11x-seg) with a P2 activation layer for small-object detection and Focaler-MPDIoU loss function was trained on 250 phase-contrast micrographs of an industrial methanotrophic consortium based on Methylococcus capsulatus KN2, cultivated continuously at dilution rates of 0.15-0.25 h-1. The dataset comprised nine cellular morphotypes across 50,410 annotated objects (training: 200 images; validation: 50 images), with synthetic data augmentation applied to reduce class imbalance in the tetracocci class. The trained model was then applied to a pure M. capsulatus KN2 culture during substrate-unlimited batch growth (μmax = 0.223 h-1, td = 3.11 h) across three biological replicates and four time points (3, 5, 7, 9 h), analyzing 277 micrographs and 15,124 objects.
RESULTS: On the validation set, the model achieved mAP@0.5:0.95 = 0.52, with class-weighted Precision = 0.87, Recall = 0.85, and F1 = 0.82. Per-morphotype F1 scores were: monococci 0.75, diplococci0.89, tetracocci 0.65. In the industrial consortium, producer cells (M. capsulatus KN2) constituted 88.5% of the population (monococci 33.7%, diplococci 61.9%, tetracocci 4.4%), while satellite bacteria comprised 11.5%. During batch cultivation of the pure producer strain, the diplococci fraction increased from 61.0% to 68.6% over 7 h, negatively correlating with a decline in monococci from 35.4% to 28.8% (Pearson r = -0.996, p = 0.004). Tetracocci showed no statistically significant correlation with either morphotype and are considered a stochastic subpopulation of diplococci. Cell cycle analysis revealed elongation of the M-phase from 71.6% to 78.4% of td. Monococci cell radius, intracellular volume, and periplasmic surface area all increased over 9 h, while the surface-to-volume ratio declined.
DISCUSSION/CONCLUSION: The observed M-phase elongation is consistent with incipient substrate limitation (CH4 or O2) in gas-tight batch flasks, detectable through morphotype ratio shifts before standard process parameters register any change. The approach enables a proof-of-concenpt for real-time culture quality monitoring, early prediction of growth limitations, and optimization of SCP production in industrial bioreactors.
METHODS: A convolutional neural network (YOLO11x-seg) with a P2 activation layer for small-object detection and Focaler-MPDIoU loss function was trained on 250 phase-contrast micrographs of an industrial methanotrophic consortium based on Methylococcus capsulatus KN2, cultivated continuously at dilution rates of 0.15-0.25 h-1. The dataset comprised nine cellular morphotypes across 50,410 annotated objects (training: 200 images; validation: 50 images), with synthetic data augmentation applied to reduce class imbalance in the tetracocci class. The trained model was then applied to a pure M. capsulatus KN2 culture during substrate-unlimited batch growth (μmax = 0.223 h-1, td = 3.11 h) across three biological replicates and four time points (3, 5, 7, 9 h), analyzing 277 micrographs and 15,124 objects.
RESULTS: On the validation set, the model achieved mAP@0.5:0.95 = 0.52, with class-weighted Precision = 0.87, Recall = 0.85, and F1 = 0.82. Per-morphotype F1 scores were: monococci 0.75, diplococci0.89, tetracocci 0.65. In the industrial consortium, producer cells (M. capsulatus KN2) constituted 88.5% of the population (monococci 33.7%, diplococci 61.9%, tetracocci 4.4%), while satellite bacteria comprised 11.5%. During batch cultivation of the pure producer strain, the diplococci fraction increased from 61.0% to 68.6% over 7 h, negatively correlating with a decline in monococci from 35.4% to 28.8% (Pearson r = -0.996, p = 0.004). Tetracocci showed no statistically significant correlation with either morphotype and are considered a stochastic subpopulation of diplococci. Cell cycle analysis revealed elongation of the M-phase from 71.6% to 78.4% of td. Monococci cell radius, intracellular volume, and periplasmic surface area all increased over 9 h, while the surface-to-volume ratio declined.
DISCUSSION/CONCLUSION: The observed M-phase elongation is consistent with incipient substrate limitation (CH4 or O2) in gas-tight batch flasks, detectable through morphotype ratio shifts before standard process parameters register any change. The approach enables a proof-of-concenpt for real-time culture quality monitoring, early prediction of growth limitations, and optimization of SCP production in industrial bioreactors.