Big data, small explanatory and predictive power: Lessons from random forest modeling of on-farm yield variability and implications for data-driven agronomy

Martin van Ittersum (2023, [Artículo])

Context: Collection and analysis of large volumes of on-farm production data are widely seen as key to understanding yield variability among farmers and improving resource-use efficiency. Objective: The aim of this study was to assess the performance of statistical and machine learning methods to explain and predict crop yield across thousands of farmers’ fields in contrasting farming systems worldwide. Methods: A large database of 10,940 field-year combinations from three countries in different stages of agricultural intensification was analyzed. Random effects models were used to partition crop yield variability and random forest models were used to explain and predict crop yield within a cross-validation scheme with data re-sampling over space and time. Results: Yield variability in relative terms was smallest for wheat and barley in the Netherlands and for wheat in Ethiopia, intermediate for rice in the Philippines, and greatest for maize in Ethiopia. Random forest models comprising a total of 87 variables explained a maximum of 65 % of cereal yield variability in the Netherlands and less than 45 % of cereal yield variability in Ethiopia and in the Philippines. Crop management related variables were important to explain and predict cereal yields in Ethiopia, while predictive (i.e., known before the growing season) climatic variables and explanatory (i.e., known during or after the growing season) climatic variables were most important to explain and predict cereal yield variability in the Philippines and in the Netherlands, respectively. Finally, model cross-validation for regions or years not seen during model training reduced the R2 considerably for most crop x country combinations, while for wheat in the Netherlands this was model dependent. Conclusion: Big data from farmers’ fields is useful to explain on-farm yield variability to some extent, but not to predict it across time and space. Significance: The results call for moderate expectations towards big data and machine learning in agronomic studies, particularly for smallholder farms in the tropics where model performance was poorest independently of the variables considered and the cross-validation scheme used.

Model Accuracy Model Precision Linear Mixed Models CIENCIAS AGROPECUARIAS Y BIOTECNOLOGÍA MACHINE LEARNING SUSTAINABLE INTENSIFICATION BIG DATA YIELDS MODELS AGRONOMY

Statistical machine-learning methods for genomic prediction using the SKM library

Osval Antonio Montesinos-Lopez Brandon Alejandro Mosqueda González Jose Crossa (2023, [Artículo])

Sparse Kernel Methods R package Statistical Machine Learning Genomic Selection CIENCIAS AGROPECUARIAS Y BIOTECNOLOGÍA MARKER-ASSISTED SELECTION MACHINE LEARNING GENOMICS METHODS

High-throughput plot-level quantitative phenotyping using convolutional neural networks on very high-resolution satellite images

Francisco Pinto Matthew Paul Reynolds Robert Furbank (2024, [Artículo])

Deep Learning Object-Based Image Analysis Optical Imagery CIENCIAS AGROPECUARIAS Y BIOTECNOLOGÍA AGRICULTURE IMAGE ANALYSIS PLANT BREEDING REMOTE SENSING MACHINE LEARNING

Multimodal deep learning methods enhance genomic prediction of wheat breeding

Carolina Rivera-Amado Francisco Pinto Francisco Javier Pinera-Chavez David González-Diéguez Matthew Paul Reynolds Paulino Pérez-Rodríguez Huihui Li Osval Antonio Montesinos-Lopez Jose Crossa (2023, [Artículo])

Conventional Methods Genomic Prediction Accuracy Deep Learning Novel Methods CIENCIAS AGROPECUARIAS Y BIOTECNOLOGÍA WHEAT BREEDING MACHINE LEARNING METHODS MARKER-ASSISTED SELECTION

Interfaces hápticas para identificar marcadores digitales del Trastorno del Espectro Autista

Haptic interfaces to identify digital markers of Autism Spectrum Disorder

GLORIA IVONNE MONARCA PINTLE (2023, [Tesis de doctorado])

El uso de herramientas de cribado durante la infancia es de suma importancia para identificar a niños que pueden Trastorno del Espectro Autista (TEA); sin embargo, las pruebas que existen en la actualidad se basan en cuestionarios y observaciones contestadas por los padres por lo que el resultado puede ser subjetivo. Además, la mayoría de las pruebas de cribado se enfocan en el área social y de comunicación, dejando a un lado las características sensoriales, un área donde los padres notan las primeras señales de un desarrollo atípico. En los últimos años, ha crecido el interés por identificar marcadores digitales del TEA que puedan apoyar el cribado de dicho trastorno, usando tecnología que pueda ser usada en cualquier lugar y momento para agilizar el proceso de cribado. Hasta el momento, la investigación en torno a marcadores digitales del TEA relacionados con el procesamiento táctil, ha sido poco explorada debido a que la investigación se ha centrado en explorar otros comportamientos, como la atención visual, movimientos motores, y voz. El crecimiento de las interfaces hápticas abre la posibilidad de explorar el procesamiento táctil e identificar marcadores digitales del TEA. En esta tesis exploramos si las interfaces hápticas pueden ser usadas para promover y recolectar interacciones táctiles que revelen marcadores digitales del TEA. Para tener un mejor entendimiento de las interfaces hápticas, en este trabajo de tesis exploramos dos interfaces hápticas. Una interfaz háptica pasiva llamada BendableSound, la cual cuenta con una retroalimentación por textura y espacio; y una interfaz háptica activa llamada Feel and Touch, la cual es un juego móvil que utiliza patrones vibrotactiles para retroalimentar las interacciones de los niños. Los resultados muestran que las interfaces hápticas tienen la capacidad recolectar interacciones táctiles y revelar marcadores digitales que pueden estar asociados al TEA. En el caso pasivo, participaron 37 niños neurotípicos (NT) y 22 niños con TEA quienes interactuar con BendableSound; los resultados muestran que existen diferencias entre los niños NT y TEA en cuanto al ancho de las interacciones táctiles, la fuerza que aplican y el tiempo que utilizan para realizar interacciones táctiles. En el caso activo, participaron 36 niños NT y 19 niños con TEA quienes interactuaron con Feel and Touch; los resultados muestran que existen diferencias entre los niños NT y TEA en cuanto a la ...

The use of screening tools during infancy is crucial to identify children who are likely to have Autism Spectrum Disorder (ASD); however, the current screening tests are based on questionnaires and observations that parents answer bases on their observations; however, the results might be subjective. In addition, most screening tests focus on the assessment of social and communication skills, disregard sensory skills, an area where parents often notice the first signals of ASD. In recent years, there has been growing interest in identifying digital markers of ASD that can support screening, particularly those that that can be used at any time and any place to speed up the screening process. So far, research on digital markers of ASD related to tactile processing has been little explored due to the lack of tools to assess it. The growth of haptic interfaces opens the possibility to explore tactile processing and identify digital markers of ASD. In this thesis, we propose to investigate the use of haptic interfaces to promote and collect tactile interactions that can reveal digital markers of ASD. To have a better understanding of haptic interfaces, in this thesis work we explore two types of haptic interface. A passive haptic interface called BendableSound, which features texture and spatial feedback; and an active haptic interface called Feel and Touch, which is a mobile game that uses vibrotactile patterns to provide feedback on children’s interactions. The results show that haptic interfaces have the potential to collect touch interactions and reveal digital markers that may be associated with ASD. In the first case, 37 neurotypical (NT) children and 22 children with ASD participated and interacted with BendableSound; the results show that there are differences between NT and ASD children in the width of tactile interactions, the force they apply, and the time they use to perform tactile interactions. In the second case, 36 NT children and 19 children with ASD participated and interacted with Feel and Touch; the results show that there are differences between NT and ASD children in terms of the tilt they produce on the mobile when performing tactile interactions, the total acceleration they produce, and the closeness to the center of the screen of their tactile interactions. The results of this thesis indicate that haptic interfaces have the ability to gather touch interactions and uncover digital markers associated with ASD. The findings of this ...

autismo, cribado, marcadores digitales, interfaces hápticas, aprendizaje automático autism, screnning, digital markers, haptic interfaces, machine learning INGENIERÍA Y TECNOLOGÍA CIENCIAS TECNOLÓGICAS TECNOLOGÍA DE LOS ORDENADORES ORDENADORES DIGITALES ORDENADORES DIGITALES

Predicción de actividad antimicrobiana usando modelos de escala evolutiva a través de un flujo de trabajo en la plataforma KNIME

Prediction of antimicrobial activity using models of evolutionary scale through a workflow in the KNIME platform

Karla Lorena MartÍnez Mauricio (2023, [Tesis de maestría])

Dentro de las estrategias para combatir la resistencia antimicrobiana, se está llevando a cabo investigación para la creación de nuevos fármacos basados en péptidos antimicrobianos. En los últimos años, se han realizado esfuerzos para incorporar herramientas computacionales que ayuden a acelerar la identificación de péptidos con actividad antimicrobiana. Una de estas herramientas son los modelos QSAR basados en aprendizaje tradicional, que permiten predecir la actividad antimicrobiana en péptidos a partir de información basada en su secuencia. Un componente clave en este proceso es el tipo de características moleculares a utilizar. Recientemente, ha surgido una familia de modelos pre-entrenados llamados ESM-2, los cuales generan incrustaciones (características) que fueron aprendidas a partir de 65 millones de secuencias que abarcan diversidad evolutiva. En este trabajo de tesis, se analiza la contribución de las incrustaciones ESM-2 de diferentes dimensiones de forma individual y en conjunto en el desarrollo de modelos QSAR basados en aprendizaje tradicional para la clasificación de péptidos antimicrobianos, así como sus tipos funcionales, como antibacteriano, antifúngico y antiviral. A partir de este estudio se concluye que aumentar la capacidad de los modelos ESM-2 no implica una mejora en el rendimiento de los modelos para predecir péptidos antimicrobianos. Los modelos ESM-2 t30 y ESM-2 t33 son los más apropiados para extraer características y mejorar la exactitud en las predicciones de péptidos antimicrobianos. Además, fusionar características de diferentes incrustaciones ESM-2 es una estrategia efectiva para construir mejores modelos QSAR que el uso exclusivo de características derivadas de un modelo ESM-2 específico. Se construyeron modelos más simples con un rendimiento comparable o superior a los modelos basados en aprendizaje profundo reportados en la literatura. Para llevar a cabo este estudio se implementó un flujo de trabajo en KNIME que genera de forma automática hasta 1980 modelos de clasificación binaria basados en aprendizaje tradicional. Incorpora diversas técnicas de selección de características, algoritmos de clasificación, métricas de desempeño y una fase de limpieza de datos. Este flujo de trabajo se encuentra disponible en https://github.com/cicese-biocom/classification-QSAR-bioKom.

Molecular features play an important role in different bio-chem-informatics tasks, such as the Quantitative Structure-Activity Relationships (QSAR) modeling. Several pre-trained models have been recently created to be used in downstream tasks either by fine-tuning a specific model or by extracting features to feed traditional classifiers. In this sense, a new family of Evolutionary Scale Modeling models (termed as ESM-2 models) has been recently introduced, demonstrating outstanding results in structure protein prediction benchmarks. Herein, we are devoted to assessing the usefulness of different-dimensional embeddings derived from ESM-2 models in the prediction of antimicrobial peptides, given the great deal of attention received because of their potential to become a plausible option to mainly fight multi-drug resistant bacteria. To this end, we created a KNIME workflow to guarantee using the same modeling methodology, and consequently, carrying out fair comparisons. As a result, it can be drawn that the 640- and 1,280- dimensional embeddings are the most appropriate to be used in modeling because statistically better results were achieved from them. We also combined features from different embeddings, and we can draw that the fusion of features of different embeddings contributes to getting better models than only using a specific model ESM-2. Comparisons regarding state-of-the-art deep learning models confirm that when performing methodologically principled studies in the prediction of AMPs, non-DL based models yield comparable-to-superior results to DL-based models. The implemented KNIME workflow is availablefreely at https://github.com/cicese-biocom/classification-QSAR-bioKom. We consider that this workflow can be valuable to prevent unfair comparisons regarding new computational methods, as well as to propose new non-DL based models.

péptidos antimicrobianos, QSAR, aprendizaje automático ESM-2, KNIME antimicrobial peptides, QSAR, machine learning, ESM-2, KNIME INGENIERÍA Y TECNOLOGÍA CIENCIAS TECNOLÓGICAS TECNOLOGÍA DE LOS ORDENADORES DISEÑO CON AYUDA DE ORDENADOR DISEÑO CON AYUDA DE ORDENADOR

Contratos inteligentes para la gestión de datos de sensado móvil y vestible para aplicaciones en salud

Smart contracts for mobile and wearable sensing data management for health applications

José Ricardo Cedeño García (2023, [Tesis de maestría])

El aumento en la producción de datos derivado de la adopción de tecnologías móviles y de IoT está revolucionando la salud, pero también plantea importantes retos éticos y de privacidad. Los recientes avances en el aprendizaje automático han resaltado la importancia de recopilar y etiquetar datos correctamente, en especial para fines críticos, como el desarrollo de aplicaciones para cuidados médicos. La recopilación de datos médicos para tareas de aprendizaje automático presenta limitaciones en cuanto a la cantidad, variedad y calidad de las fuentes disponibles. Una forma de abordar este dilema es el uso de Blockchain para la recopilación y el uso de datos de pacientes. El anonimato de una red centralizada permite proteger la identidad del paciente. La estructura formada por nodos permite que la información esté siempre disponible y no dependa de un servidor principal. La inmutabilidad de los registros en la cadena garantiza la trazabilidad inequívoca del flujo de los datos del paciente. Por último, los mecanismos de consenso y recompensa de la red podrían motivar a nuevos usuarios a participar del sensado activo. Presentamos TRHEAD, una arquitectura de referencia basada en la Blockchain para recopilar datos sanitarios, firmar consentimientos, anotar datos y obtener crédito por los mismos, permitiendo a los usuarios rastrear el uso de sus datos, a los científicos rastrear su procedencia y proteger al mismo tiempo la privacidad de los pacientes. Exponemos dos implementaciones de nuestra arquitectura aplicadas a distintas campañas de sensado para comprobar su viabilidad, así como los resultados de su aplicación en estos escenarios y las conclusiones que desprendieron de su análisis. Dado que uno de los objetivos principales de TRHEAD es la recopilación de datos mediante sensado activo para el entrenamiento legal/consciente de modelos de aprendizaje automático, se realizó el entrenamiento de un modelo con los datos obtenidos de la campaña de sensado correspondiente a imágenes de rostros humanos, con el fin de detectar estados de ánimo. Finalmente se discute el papel de TRHEAD en el aseguramiento del trato justo y consciente de la información de los pacientes y el camino por recorrer en el perfeccionamiento de la arquitectura.

The increase in data production resulting from the adoption of mobile and IoT technologies is revolutionizing healthcare, but it also poses significant ethical and privacy challenges. Recent advances in machine learning have highlighted the importance of collecting and labeling data correctly, especially for critical purposes such as deploying healthcare software. Collecting medical data for machine learning tasks presents limitations in terms of the quantity, variety, and quality of available sources. One way to address this dilemma is the use of Blockchain for the collection and use of patient data. The anonymity of a centralized network allows the patient’s identity to be protected. The structure formed by nodes allows information to be always available and not dependent on a main server. The immutability of the records in the chain guarantees the unequivocal traceability of the flow of patient data. Finally, the network’s consensus and reward mechanisms could motivate new users to participate in active sensing. We present TRHEAD, a Blockchain-based reference architecture for collecting healthcare data, signing consents, annotating data and getting credit for it, allowing users to track the use of their data, scientists to track its provenance while protecting patients privacy. We present two implementations of our architecture applied to different sensing campaigns to test their feasibility, as well as the results of their application in these scenarios and the conclusions drawn from those results. Since one of the main objectives of TRHEAD is the collection of data through active sensing for the legal/conscious training of machine learning models, a model was trained with the data obtained from the sensing campaign corresponding to images of human faces, in order to detect moods. Finally, the role of TRHEAD in ensuring the fair and conscientious treatment of patient information and the road ahead in refining the architecture is discussed.

Contratos Inteligentes, Blockchain, Privacidad, Aprendizaje de Máquina Etico, Recopilación Consciente de Datos, Consentimiento, Arquitectura de Referencia Smart Contracts, Blockchain, Privacy, Ethical Machine Learning, Conscious Data Collection, Consent, Reference Architecture INGENIERÍA Y TECNOLOGÍA CIENCIAS TECNOLÓGICAS TECNOLOGÍA DE LOS ORDENADORES INTELIGENCIA ARTIFICIAL INTELIGENCIA ARTIFICIAL

Situated learning through local stories: positioning socio-ecological concerns, knowledge and practices in school

Rosa Guadalupe Mendoza Zuany Juan Carlos A. Sandoval Rivera Paula Martínez Bautista (2023, [Artículo, Artículo])

The article analyzes the importance of the local stories, which contain concerns, knowledge and practices to look after the socio-ecological environment, to trigger situated and pertinent teaching and learning processes in primary education in rural and indigenous contexts in Veracruz, Mexico. In particular, it focuses on the stories about care in the framework of a deep socio-ecological crisis in two Nahua communities of the Huasteca region. Interview-conversations were carried out with community actors in the two communities in which 32 stories emerged. The analysis allowed the identification of socio-ecological concerns of the community, characteristics of the stories that show their potential in the learning processes, as well as types of knowledge and practices that are rarely considered in the classroom, which are capable of being linked to curricular content, in order to contribute to reflection and action on the socio-ecological crisis of the communities.

aprendizaje narración de historias medio ambiente ambiente socio-cultural escuela rural CIENCIAS SOCIALES; HUMANIDADES Y CIENCIAS DE LA CONDUCTA CIENCIAS SOCIALES HUMANIDADES Y CIENCIAS DE LA CONDUCTA Learning story telling environment socio-cultural environment rural school

COVID-19 and family participation in school activities.: Teaching experiences around the Learn at Home program

Evangelina Cervantes Holguín Pavel Roel Gutiérrez Sandoval (2022, [Artículo, Artículo])

The article analyzes, from the qualitative method, the participation of families and teaching staff of the first cycle of primary education in the state of Chihuahua (Mexico) to carry out the various activities of the Learn at Home program implemented in March 2020 as a response to the resulting health confinement by COVID-19. It is concluded that the participation of families in school emergency situations implies improving the relationship between teachers, families, and the community in the implementation processes of educational programs with greater support to organize study times, take advantage of the different cultural capitals and promote family co-responsibility.

Aprendizaje Educación a distancia Enseñanza primaria Epidemia Familia COVID-19 Chihuahua HUMANIDADES Y CIENCIAS DE LA CONDUCTA HUMANIDADES Y CIENCIAS DE LA CONDUCTA Family Elementary education Epidemics Distance education Learning

Multi-modal deep learning improves grain yield prediction in wheat breeding by fusing genomics and phenomics

Xu Wang Sandesh Kumar Shrestha Philomin Juliana Suchismita Mondal Francisco Pinto Govindan Velu Leonardo Abdiel Crespo Herrera JULIO HUERTA_ESPINO Ravi Singh Jesse Poland (2023, [Artículo])

New Crop Varieties Plant Breeding Programs Yield Prediction CIENCIAS AGROPECUARIAS Y BIOTECNOLOGÍA LEARNING GRAIN YIELDS WHEAT BREEDING FOOD SECURITY