Mosquito-borne diseases like malaria are rising globally, and improved mosquito vector surveillance is needed. Survival of Anopheles mosquitoes is key for epidemiological monitoring of malaria transmission and evaluation of vector control strategies targeting mosquito longevity, as the risk of pathogen transmission increases with mosquito age. However, the available tools to estimate field mosquito age are often approximate and time-consuming. Here, we show a rapid method that combines matrix-assisted laser desorption/ionization–time-of-flight mass spectrometry with deep learning for mosquito age prediction. Using 2763 mass spectra from the head, legs, and thorax of 251 field-collected Anopheles arabiensis mosquitoes, we developed deep learning models that achieved a best mean absolute error of 1.74 days. We also demonstrate consistent performance at two ecological sites in Senegal, supported by age-related protein changes. Our approach is promising for malaria control and the field of vector biology, benefiting other disease vectors like Aedes mosquitoes.

Identifying fungal clones propagated during outbreaks in hospital settings is a problem that increasingly confronts biologists. Current tools based on DNA sequencing or microsatellite analysis require specific manipulations that are difficult to implement in the context of routine diagnosis. Using deep learning to classify the mass spectra obtained during the routine identification of fungi by MALDI-TOF mass spectrometry could be of interest to differentiate isolates belonging to epidemic clones from others. As part of the management of a nosocomial outbreak due to Candida parapsilosis in two Parisian hospitals, we studied the impact of the preparation of the spectra on the performance of a deep neural network. Our purpose was to differentiate 39 otherwise fluconazole-resistant isolates belonging to a clonal subset from 56 other isolates, most of which were fluconazole-susceptible, collected during the same period and not belonging to the clonal subset. Our study carried out on spectra obtained on four different machines from isolates cultured for 24 or 48 h on three different culture media showed that each of these parameters had a significant impact on the performance of the classifier. In particular, using different culture times between learning and testing steps could lead to a collapse in the accuracy of the predictions. On the other hand, including spectra obtained after 24 and 48 h of growth during the learning step restored the good results. Finally, we showed that the deleterious effect of the device variability used for learning and testing could be largely improved by including a spectra alignment step during preprocessing before submitting them to the neural network. Taken together, these experiments show the great potential of deep learning models to identify spectra of specific clones, providing that crucial parameters are controlled during both culture and preparation steps before submitting spectra to a classifier.

Mycobacterium abscessus (MABS) displays differential subspecies susceptibility to macrolides. Thus, identifying MABS's subspecies ( M. abscessus , M. bolletii and M. massiliense ) is a clinical necessity for guiding treatment decisions. We aimed to assess the potential of Machine Learning (ML)‐based classifiers coupled to Matrix‐Assisted Laser Desorption/Ionization Time‐of‐Flight (MALDI‐TOF) MS to identify MABS subspecies. Two spectral databases were created by using 40 confirmed MABS strains. Spectra were obtained by using MALDI‐TOF MS from strains cultivated on solid (Columbia Blood Agar, CBA) or liquid (MGIT®) media for 1 to 13 days. Each database was divided into a dataset for ML‐based pipeline development and a dataset to assess the performance. An in‐house programme was developed to identify discriminant peaks specific to each subspecies. The peak‐based approach successfully distinguished M. massiliense from the other subspecies for strains grown on CBA. The ML approach achieved 100% accuracy for subspecies identification on CBA, falling to 77.5% on MGIT®. This study validates the usefulness of ML, in particular the Random Forest algorithm, to discriminate MABS subspecies by MALDI‐TOF MS. However, identification in MGIT®, a medium largely used in mycobacteriology laboratories, is not yet reliable and should be a development priority.

The ongoing worldwide emergence of epidemic azole-resistant Candida parapsilosis clones is a threat to human health. In order to monitor a clonal outbreak which involved 2 hospitals in Paris, we used a technology that most clinical laboratories are equipped with: Matrix-Assisted Laser Desorption-Ionization Time of Flight mass spectrometry (MALDI-TOF MS). This physico-chemical analysis technique was combined with machine learning algorithm to evaluate the possibility of monitoring a multi-center epidemic. A total of 96 isolates of Candida parapsilosis constituted of resistant clones, susceptible clones and other susceptible isolates were cultured on various culture conditions. Spectra were acquired on four different mass spectrometers from three hospitals. We used convolutional neural networks (CNN) to classify the isolates and identified the main parameters that can influence the identification of epidemic clonal isolates. We have shown results up to 94% accuracy in a multi-center context, and between 81 and 91% accuracy when the tested site is not part of the training set, showing that it is possible to differentiate the spectra of a clone and a non-clone presenting a very strong similarity. This study shows that it is possible to monitor a certain type of clone, namely the R2 clones in our case study, using MALDI-TOF coupled to CNN. The experimental conditions that impacted the results should be considered for future work to improve performances for difficult classification tasks like clonal fungal identification.

La spectrométrie de masse de type MALDI-TOF (matrix assisted laser desorption and ionisation time of flight) est une méthode de diagnostic en microbiologie rapide et robuste, permettant d'identifier les espèces de micro-organismes grâce à leur empreinte protéique constituée par le spectre de masse. Cependant, les applications clinico-épidémiologiques de cette technologie demeurent limitées par les outils bio-informatiques à disposition. Cette thèse se focalise sur l'application de modèles d'apprentissage statistique profonds aux données de spectrométrie de masse de type MALDI-TOF dans un but de surveillance épidémiologique des maladies infectieuses. Elle inclut la surveillance des épidémies de champignons et de mycobactéries en milieu hospitalier, ainsi que la caractérisation des anophèles vecteurs du paludisme. Nous avons examiné l'impact des méthodes de préparation des échantillons et de l'analyse informatique des spectres de masse sur l'amélioration de l'apprentissage, afin d’identifier les clones fongiques épidémiques en milieu hospitalier et prévenir leur propagation. Notre étude a montré que le réseau de neurones à convolution (CNN) a un potentiel élevé pour identifier les spectres de clones spécifiques de Candida parapsilosis, atteignant une précision de 94 % en optimisant des paramètres essentiels (milieux de culture, temps de croissance, et la machine d'acquisition des spectres). Pour détecter des clones épidémiques Aspergillus flavus dans des cohortes hospitalières multicentriques, le CNN a également réussi à classer correctement la plupart des isolats, atteignant une précision supérieure à 93 % pour deux des trois appareils utilisés. Nous avons aussi montré qu’en utilisant des modèles d'apprentissage profond optimisés, tels qu'un CNN et un réseau de neurones à convolution temporelle (TCN), nous pouvons prédire l'âge des moustiques avec une précision moyenne de deux jours (meilleure erreur absolue moyenne : 1,74 jours). Cette approche permettrait ainsi de surveiller efficacement la structure de l'âge des populations de moustiques anophèles sauvages et de mieux les cibler par des mesures de contrôle. Enfin, nous avons démontré les performances de diverses architectures de réseaux de neurones et de différentes méthodes de représentation des spectres de masse, en utilisant différentes cohortes couvrant diverses problématiques épidémiologiques telles que la prédiction de l'âge, l'identification d'espèces étroitement apparentées des moustiques anophèles, la distinction entre sous-espèces proches, ainsi que la détection de la résistance chez le Mycobacterium abscessus. L'étude a montré que parmi les différents modèles évalués, les modèles les plus performants, tels que les TCN et un réseau de neurones récurrents, pouvaient obtenir des résultats notables, atteignant une précision d'identification de 93 % pour les espèces d'Anophèles étroitement liées et de 95 % pour les sous-espèces de Mycobacterium abscessus. De plus, l'utilisation de CNN et de TCN a permis de détecter les souches résistantes chez Mycobacterium abscessus avec une précision dépassant 97 %. Cette thèse met en lumière l'utilisation de l'apprentissage profond en conjonction avec le MALDI-TOF, une approche jusqu'ici peu explorée. Avec la généralisation des instruments MALDI-TOF et la possibilité de coupler les analyses à des applications en ligne utilisant l'apprentissage profond, cette approche semble prometteuse, ouvrant la voie à d'autres applications épidémiologiques au-delà de la simple identification d’espèce, telles que la détection de clusters épidémiologiques de microorganismes résistants aux médicaments, la surveillance de la transmission des maladies bactériennes et fongiques, et l’évaluation de l'efficacité des interventions ciblées de lutte antivectorielle.

Emerging global health threats, from antimicrobial resistance to vector-borne diseases, require scalable diagnostic solutions. Matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) mass spectrometry has revolutionized microbial diagnostics but remains underutilized in public health due to algorithmic challenges in interpreting and generalizing complex spectra. Here, we present a systematic benchmarking framework to expand MALDI-TOF's utility for field-deployable diagnostics using diverse deep learning architectures and preprocessing strategies, including approaches newly applied to spectral data. Using a total of 7424 mass spectra from clinical and field-collected datasets, we evaluated Mycobacterium abscessus subspecies and antimicrobial resistance classification (814 training/187 test spectra from 33/8 isolates); Anopheles mosquito species identification across anatomical parts (legs: 1047/188 spectra from 211/40 specimens; head: 1089/214 spectra; thorax: 999/279 spectra); and mosquito age estimation (legs: 806/144 spectra from 202/36 specimens; head: 727/124 spectra; thorax: 689/117 spectra). Results demonstrate that balanced accuracy ranged from 84% to 95% for species/subspecies identification and from 91% to 99% for resistance prediction, and that mean absolute error ranged from 1.9 to 3.8 days for age estimation. Key to this performance are adaptations managing instrumental and biological variability. We further assess computational efficiency and energy consumption, highlighting trade-offs between accuracy and deployment feasibility. These findings support AI-enhanced MALDI-TOF as a practical tool for global health surveillance and precision diagnostics in resource-limited settings.