In the pursuit of effective control of malaria, especially in regions such as sub-Saharan Africa where the burden of the disease is highest, the development of rapid, cost-effective tools to monitor transmission dynamics is essential and urgent. Malaria parasites-Infectious Anopheles mosquito This is especially important for understanding transmission patterns in different regions, estimating the impact of interventions, and planning new interventions. Unfortunately, current detection methods, mainly ELISA and PCR, do not allow detection of HIV-1 infections. Malaria parasites in Anopheles mosquito Mosquito control is resource intensive, requiring specialised skills and supplies that are often lacking locally – limitations that make detailed assessments of malaria risk and the effectiveness of interventions at the district level difficult.
Our work presents a new and economical approach combining mid-infrared (MIR) spectroscopy and supervised machine learning algorithms for rapid identification. Malaria parasites-Infectious Anopheles mosquito By collecting and analyzing mid-infrared spectra from the head and thorax of wild-caught mosquitoes, Hey, Funestus We validated these findings by ELISA or PCR in rural women in Tanzania. Plasmodium falciparum By modeling sporozoites, we established a reliable “ground truth” for model training. The results of this study are compelling, demonstrating that MIR spectral analysis can distinguish infectious from non-infectious mosquitoes with over 90% accuracy. Notably, models trained on PCR data generalized better compared to models based on ELISA data, and mosquito age did not have a significant interference. Plasmodium falciparum and Hey, FunestusThis advance marks an important step in malaria surveillance. If adapted for other major African tropical malaria vectors and malaria transmission systems, it could provide a scalable, low-cost solution that could transform data-driven decision-making in disease control programs. Moreover, while we see this as an important step toward building a system that can be deployed immediately, we acknowledge that further development is required. Models trained with more diverse data from different settings will improve observational accuracy and strengthen readiness for more widespread deployment of this approach.
This study contributes to expanding knowledge demonstrating the potential of MIRS-ML-based approaches in malaria vector surveillance. The use of these methodologies to delineate key entomological parameters such as mosquito age, species identification and blood-feeding patterns has been well documented.21,22,25“The results of our study suggest that this technology can serve as a multi-purpose platform, whereby infrared scans can be interpreted to determine not only the species and age of a mosquito, which are important factors in its potential as a malaria vector, but also its blood-feeding history on humans and other vertebrates, and its infection status with the malaria parasite. Such comprehensive profiling will help to accurately characterize malaria risk, and represent a major advancement for vector surveillance and malaria control strategies.”
In addition to the high classification accuracy of the MIRS-ML approach, the PCR-trained model achieved a generalization rate of over 85% in predicting sporozoite infection in wild-caught cattle. Hey, Funestus Mosquitoes can be identified even when predicting the results of an ELISA dataset. These results achieve consistent performance with studies using NIRS frequencies in the laboratory, reporting classification accuracies of over 90% for detection. Plasmodium falciparum Sporozoite infection Gambiae mosquito15,The detection accuracy is 77%. P. Bergey Sporozoite infection Anatolia16Previous models trained with NIRS have been unable to distinguish mosquitoes infected with wild-type parasites from asymptomatic malaria carriers, likely due to limitations in the training dataset or the detection capabilities of the systems.33Models trained with MIRS, which provides sharper peaks and richer biochemical information, appear to perform better.18,19This enhancement enabled our model to effectively identify mosquito infections, a capability that had not been fully realized by NIRS models in previous studies.
MIRS captures the biochemical composition of the mosquito, which in this case may always vary depending on the infection state, such as the presence or absence of parasites. The presence of parasite-specific proteins, such as circumsporozoite (CS) proteins and thrombospondin-related adhesion proteins (TRAPs), may contribute to the main spectrum differences between infectious and non-infectious mosquitoes. Hey, Funestus34Furthermore, mosquitoes mount an immune response against the parasite, which may affect the biochemical properties of infectious or non-infectious mosquitoes.35Furthermore, non-infectious mosquitoes have higher reserve levels of energy resources, including glucose and lipid accumulation.36,37May produce different spectral signals between infectious and non-infectious Hey, FunestusThis is because the majority of the spectral features that influence machine learning predictions are primarily O–H, C–H, and N–H bonds, as well as the spectral fingerprint region (1500 cm−1 Up to 520cm−1), suggesting the presence of carbohydrates, proteins, and lipids associated with the parasite.20However, it is important to note that rather than focusing on individual spectral features, we used ML models to integrate a set of spectral features from different biochemical group regions to enable these classifications. Although identifying specific features may not be essential, we believe that additional research should be conducted to better understand the biochemical signals underlying the algorithmic classification.
A biological prerequisite for becoming a malaria vector is that a certain age threshold (e.g., 9 days or more) must be exceeded due to the necessary extrinsic incubation period of the parasite.38may introduce age-related bias in detection efficiency. In this context, mosquito age can be considered a confounding factor affecting prediction accuracy. However, despite the theoretical possibility that age may affect prediction accuracy, our analysis showed that the machine learning model successfully identified signals indicative of infection in all age groups, including mosquitoes older than 14 days, denying that age is an important confounding variable in our study.
Furthermore, the ML model trained on the PCR infection dataset showed the ability to generalize predictions to samples screened by ELISA. In contrast, the model trained on the ELISA infection dataset had some limitations in predicting samples screened by PCR. Furthermore, similarities were observed in the fingerprint regions where both the ELISA and PCR models detected signals, indicating concordance in parasite detection between the two models (Fig. 3). However, clear differences were observed in the signals detected by both the ELISA and PCR models, especially in the frequencies ranging from 3500 to 3000 cm.−1Moreover, it is still unclear why the ML models are picking up different signals from this region.In addition, the generalizability of the ML models trained on the PCR infection dataset may be due to the sensitivity of PCR to detect even low numbers of mosquito sporozoites.39Leveraging the sensitivity of PCR can improve the performance of the MIRS-ML model. However, in the study by Hendershot et al.., Mosquito infection was observed 0.5–1 day after infection, indicating that PCR tests may give a positive result even when sporulation has not yet begun, potentially giving false-positive results.39This situation is more likely to lead to a positive result due to the presence of gametocytes in the mosquito's head and thorax if infectious blood has not been fully transferred to the mosquito's abdomen.
The main objective of this investigation was to demonstrate the field application of the MIRS-ML technique for malaria vector sporozoite detection and not to directly compare it with PCR or ELISA methods, which were only used to provide a reference label for training the ML model. Furthermore, this study highlights that it is only the first demonstration of the field application of the MIRS-ML technique for malaria vector sporozoite detection and requires further validation before integration into surveillance and national malaria control. Our analysis also Hey, Funestus Mosquitoes have a relatively high sporozoite rate in this region, highlighting the need to expand future models to include more vector species. We acknowledge that extending the MIRS-ML approach to all important mosquito species will require compiling a comprehensive dataset of mosquito infection spectrum, a task that will be logistically challenging, especially in fields where the prevalence of natural infection is very low. A promising solution is to employ transfer learning and integrate laboratory-generated data with field-collected samples to increase the robustness of the model.24,25In this method, models initially trained on laboratory data can be refined with new field data, facilitating the development of effective tools for field infection prediction. Furthermore, in low-transmission environments where sporozoite infection rates are low, ELISA and PCR can be used for mosquito pool testing, reducing operational costs compared to individual mosquito testing. However, the feasibility of using MIRS-ML for mosquito pool testing is unclear and needs to be investigated in the next step. Furthermore, this study shows that MIRS-ML can be used to predict the presence of mosquitoes that are highly susceptible to sporozoite infection. Malaria parasites Future research should address this issue to improve its usefulness in areas where multiple species coexist. Malaria parasites is widespread.
MIRS-ML is cost-effective as it eliminates the need for repeated reagent costs for mosquito sample testing. The only cost incurred is an initial cost of GBP 25,000 for the purchase of the FT-IR spectrometer. The FT-IR spectrometer, a portable bench-top design capable of processing approximately 60 mosquitoes per hour, has dimensions of 22 × 33 × 26 cm and needs to be plugged into an AC power source. We are currently developing an online system that will act as a centralized platform for predicting various entomological and parasitological indicators of malaria. This online system aims to facilitate the expansion of MIRS-ML and allow end users in different locations to upload spectra of unknown mosquitoes and make predictions related to infection status, species, age, or resistance status.
In conclusion, here we demonstrate the first application of mid-infrared spectroscopy combined with machine learning (MIRS-ML) for rapid and accurate detection. Plasmodium falciparum Mosquitoes collected in the field. Analysis of 7,178 females revealed Hey, Funestus In samples collected in rural Tanzania, we achieved detection accuracies of 92% and 93% against ELISA and PCR benchmarks, respectively. Moreover, EIR estimates derived from the MIRS-ML model were in close agreement with those derived from PCR and ELISA methods across low and high bite rate scenarios, demonstrating the consistency and reliability of malaria transmission predictions, allowing MIRS-ML to guide programmatic decisions on vector control. This method can process approximately 60-100 mosquitoes per hour at minimal cost, thus providing a major advancement in malaria surveillance, especially in sub-Saharan Africa where the disease has a high impact. The utility of MIRS-ML extends beyond sporozoite detection, providing insights into key entomological indicators such as mosquito age, feeding patterns, and species identification, thereby positioning it as a versatile tool in malaria risk assessment and evaluation of vector control interventions.
