Cesar Augusto Vargas-Garcia

Most quantitative models of bacterial cell division assume steady-state growth, yet natural environments are highly dynamic. Here, we infer division control parameters of E. coli from single-cell size trajectories under nonsteady growth. We describe growth as a continuous process interrupted by stochastic divisions, whose propensity depends on cell size and on the size added since birth. Combining an estimation of time-dependent single-cell growth rates with a maximum-likelihood estimator that we previously validated on synthetic data, we obtain time-resolved estimates of the mean added size at division, its noise, and a strategy parameter that measures how added size depends on birth size. This parameter vanishes for the adder, in which cells add, on average, a fixed size regardless of their size at birth, and is negative for sizer-adder control, in which larger-born cells add less size. We apply this approach to published single-cell data covering steady exponential growth, gradual nutrient depletion, and exit from starvation. We implement a time-resolved model selection based on the Akaike information criterion, concluding that the adder is sufficient during steady growth, whereas size dynamics during entry into and exit from stationary phase are better explained by transient deviations toward sizer-adder control. The framework provides a general method to quantify time-resolved changes in the scale, stochasticity, and strategy of division in nonsteady environments.

Treinta años de investigación e innovación al servicio del campo colombiano se reúnen, por primera vez, en un análisis cienciométrico riguroso que mide, mapea y reflexiona sobre la generación de conocimiento de AGROSAVIA. Apoyado en bases de datos como Scopus, Web of Science, SciELO y Google Scholar, y enriquecido con métricas alternativas, este libro reconstruye la trayectoria científica y tecnológica de la corporación entre 1993 y 2022: su evolución temática, sus principales redes de coautoría nacional e internacional, sus aliados estratégicos, sus contribuciones a los Objetivos de Desarrollo Sostenible y el lugar que hoy ocupa entre los institutos nacionales de investigación agropecuaria de América Latina y el Caribe. Más allá del recuento histórico, la obra propone una mirada hacia adelante. Articula los retos de la gestión del conocimiento en I+D+i —desde los enfoques de bioeconomía, agricultura climáticamente inteligente, seguridad alimentaria, nuevos sistemas agroalimentarios e inclusión social— con las transformaciones que la ciencia abierta y la inteligencia artificial están imponiendo a la práctica investigativa. Dirigido a investigadores, gestores de ciencia, formuladores de política pública y comunidades académicas, este es un libro escrito tanto para celebrar lo logrado como para preguntarse, con honestidad y método, hacia dónde debe orientarse la generación de conocimiento agropecuario en las próximas décadas.

Automated and objective grading of cocoa (Theobroma cacao L.) fermentation remains a major challenge because the conventional cut test relies on subjective visual inspection and is difficult to scale. In this study, we develop and evaluate a deep learning pipeline for classifying cocoa bean fermentation levels from expert-annotated cut-test images acquired under controlled conditions, enabling the systematic evaluation and comparison of multiple convolutional and transformer-based architectures under consistent preprocessing, training, and evaluation protocols. The dataset comprises 4347 segmented cocoa bean images distributed across four severely imbalanced classes, namely fermented, under-fermented, slaty, and violet. Representative architectures, including EfficientNet-B0, MobileNetV3-Large, ConvNeXt-XLarge, ViT-Base, and ViT-Large, are benchmarked to analyze the effects of class imbalance, RGB versus HSV color representation, training duration, and label-space formulation. The results show that severe class imbalance strongly degrades performance in direct four-class classification. A hierarchical binary-to-multiclass strategy significantly improves balanced recognition by separating fermented from unfermented beans prior to subclass discrimination, increasing macro-F1 scores from approximately 80–83% to 89–91%. Among the evaluated models, ViT-Base emerges as the most stable architecture across experimental settings and offers the best balance between classification performance, training stability, and computational cost. Although larger models achieve slightly higher peak performance under balanced conditions, ViT-Base provides more consistent results under realistic constraints. The proposed framework enables near-real-time inference on segmented single-bean images and supports objective, reproducible, and scalable fermentation assessment. These findings demonstrate that performance in cocoa fermentation grading is determined not only by model capacity, but also by imbalance-aware label-space design and evaluation protocols aligned with real-world cut-test conditions.

Cells achieve size homeostasis by regulating their division timing based on their size, added size, and cell cycle time. Previous research under steady-state conditions demonstrated the robustness of these mechanisms. However, their dynamic responses in fluctuating environments, such as nutrient depletion due to population growth, remain challenging to fully characterize. Currently, advances in single-cell microscopy have revealed various cellular division strategies whose underlying molecular mechanisms are complex and not always available. This study introduces a novel approach to model cell size dynamics using a piecewise deterministic Markov chain framework, where cell division events are modeled as stochastic jumps determined by a division propensity dependent on both current cell size and added size since birth. We propose a three-parameter characterization for the division process: scale (target added size at division), shape (division stochasticity), and division strategy (relevance of cell size, added size, or cell cycle duration). We derive analytical formulas for the probability of division, and with this probability, we develop a maximum likelihood estimation (MLE) framework. We implement a systematic investigation of the accuracy of inference as a function of sample size. The model's performance is studied across various scenarios, including those exhibiting dynamical changes in one or more parameters, suggesting its broad applicability for analyzing new experimental data on cell size regulation in dynamic environments.

Soil organic carbon (SOC) is an essential indicator of soil fertility, health, and carbon sequestration capacity. Its proper management improves soil structure, productivity, and resilience to climate change, making rapid and reliable SOC assessment essential for sustainable agriculture. Visible and near-infrared (Vis–NIR) spectroscopy offers a non-destructive and cost-effective alternative to conventional laboratory analyses, allowing for the simultaneous estimation of multiple soil properties from a single spectrum. This study aimed to predict SOC content using machine learning techniques applied to Vis–NIR spectra of 860 soil samples collected in the Sierra Nevada de Santa Marta, Colombia. The spectra (400–2500 nm) were acquired using a NIR spectrophotometer, and the soil organic carbon (SOC) content was quantified using a wet oxidation method that employs dichromate in an acidic medium. A hybrid modeling framework combining Random Forest (RF) with support vector regression (SVR) and XGBoost was implemented. Spectral pretreatments (Savitzky–Golay first derivative, MSC, and SNV) were compared, and spectral bands were selected every 10 nm. The 30 most relevant wavelengths were identified using RF importance analysis. Data were divided into training (80%) and test (20%) subsets using stratified random sampling, and five-fold cross-validation was applied for parameter optimization and overfitting control. The RF–XGBoost (R² = 0.86) and RF–SVR (R² = 0.85) models outperformed the individual RF and SVR models (R² < 0.7). The proposed hybrid approach, optimized through features, and advanced spectral preprocessing demonstrate a robust and scalable framework for rapid prediction of SOC and sustainable soil monitoring.

Measurements of cell size dynamics have revealed phenomeno-logical principles by which individual cells control their size across diverse organisms. One of the emerging paradigms of cell size homeostasis is the adder , where the cell cycle duration is established such that the cell size increase from birth to division is independent of the newborn cell size. We provide a mechanistic formulation of the adder considering that cell size follows any arbitrary non-exponential growth law . Our results show that the main requirement to obtain an adder regardless of the growth law (the time derivative of cell size) is that cell cycle regulators are produced at a rate proportional to the growth law and cell division is triggered when these molecules reach a prescribed threshold level. Among the implications of this generalized adder, we investigate fluctuations in the proliferation of single-cell derived colonies. Considering exponential cell size growth, random fluctuations in clonal size show a transient increase and then eventually decay to zero over time (i.e., clonal populations become asymptotically more similar). In contrast, several forms of non-exponential cell size dynamics (with adder-based cell size control) yield qualitatively different results: clonal size fluctuations monotonically increase over time reaching a non-zero value. These results characterize the interplay between cell size homeostasis at the single-cell level and clonal proliferation at the population level, explaining the broad fluctuations in clonal sizes seen in barcoded human cell lines.

This contribution explores mechanisms that regulate the dynamics of single-cell size, maintaining equilibrium around a target set point. Using the formalism of Stochastic Hybrid Systems (SHS), we consider continuous exponential growth in cell size (as determined by volume/mass/surface area). This continuous-time evolution is interspersed by cell division events that occur randomly as per a given size-dependent rate, and upon division, only one of the two daughter cells is tracked. We show that a size-independent division rate does not provide cell size homeostasis, in the sense that the variance in cell size increases unboundedly over time. Next, we consider a division rate proportional to cell size that yields the adder size control observed in several bacteria – a constant size is added on average between birth and division regardless of the newborn size. For this scenario, we obtain exact formulas for the steady-state moments (mean, variance, and skewness) of cell size. Expanding the SHS model, we explore a biologically relevant scenario where the time between successive division events is further divided into multiple discrete stages with size-dependent stage transitions. Exact moment computations demonstrate that increasing the number of stages reduces cell size variability (noise). We also find formulas considering uneven size partitioning between daughters during division, and where the division rate follows a power law of the cell size leading to deviations from adder size control. This work provides a method for estimating model parameters from observed cell size distributions and enhances our understanding of mechanisms underlying cell size regulation.

Under ideal conditions, Escherichia coli cells divide after adding a fixed cell size, a strategy known as the adder. This concept applies to various microbes and is often explained as the division that occurs after a certain number of stages, associated with the accumulation of precursor proteins at a rate proportional to cell size. However, under poor media conditions, E. coli cells exhibit a different size regulation. They are smaller and follow a sizer-like division strategy where the added size is inversely proportional to the size at birth. We explore three potential causes for this deviation: degradation of the precursor protein and two models where the propensity for accumulation depends on the cell size: a nonlinear accumulation rate, and accumulation starting at a threshold size termed the commitment size. These models fit the mean trends but predict different distributions given the birth size. To quantify the precision of the models to explain the data, we used the Akaike information criterion and compared them to open datasets of slow-growing E. coli cells in different media. We found that none of the models alone can consistently explain the data. However, the degradation model better explains the division strategy when cells are larger, whereas size-related models (power-law and commitment size) account for smaller cells. Our methodology proposes a data-based method in which different mechanisms can be tested systematically.