Leishmaniasis is a tropical and subtropical disease caused by several species of protozoa of the order Kinetoplastida, family Trypanosomatidae and genus Leishmania, which affects humans and many animal species [1]. Among the main forms of the disease, visceral leishmaniasis (VL) caused by the protozoan from the Leishmania infantum (syn L. chagasi) specie is considered the most aggressive. According to estimates by the World Health Organization, between 50,000 and 90,000 humans have been affected annually [2], with fatality reaching 95% of untreated cases. The transmission of leishmaniasis occurs mainly through arthropod vectors, more specifically those of the genus Lutzomyia [3,4], which finds in dogs the main host of the protozoan in urban environments [5].

This disease dynamic created a polemic public health strategy for VL control based on identifying infected dogs, followed by their treatment or sacrifice [6] to reduce the protozoan reservoirs. Then, the need for infected dog identification demands accurate, fast, and low-cost diagnostic tests. Several tests are available for dog VL diagnosis, from molecular tests to rapid tests based on immunochromatography [7,8]. Although their wide range of advantages, we can always find some drawbacks, such as cost, high occurrence of false-positive or false-negative results, time-consuming methodology, complexity, etc. Then, there is a continuous search for more reliable methods that overcome such disadvantages in the literature, which may reach better levels of precision and accuracy, reducing the possibility of wrong results and avoiding the lead a healthy animal to sacrifice or even allowing a sick animal to continue serving as a source of infection for other animals and humans.

In the past few years, many studies have been devoted to the use of optical spectroscopy with machine learning for disease diagnosis. Since biofluids (blood serum, plasma, saliva, urine, etc.) undergo major changes in chemical and biochemical properties due to different immunological responses, such differences can be accessed by optical spectroscopy techniques, identified and used for diagnostic purposes. Infectious [9], [10], [11] or non-infectious diseases [12,13] have been successfully diagnosed by using different types of optical spectroscopy such as Raman, infrared, fluorescence, and ultraviolet-visible (UV–vis), with the aid of machine learning algorithms [14], [15], [16]. Since the main differences observed in the biofluids are assigned to molecular changes Raman and infrared spectroscopy, have been extensively explored for new diagnostic methods. But UV–Vis spectroscopy can also be applied to access molecular information, it is a common and routine technique used in microbiology as a tool to check different types of samples, performing qualitative and quantitative characterization. Typically, modern UV–Vis spectrophotometers can scan in the region of approximately 200–1000 nm in the electromagnetic spectrum, where electronic transitions in molecular orbitals assigned to important proteins and amino acids are observed [16].

UV–vis spectroscopy has already been applied to determine protein concentration in plasma serum [17], which has been used for the diagnosis of canine cancer [16], tuberculous meningitis [18], leukemia [14], and even the identification of bacteria strains [19]. Such studies are usually based on univariate analysis. Still, the information accessed by UV–vis spectroscopy is related to electronic transition bands, which may not suffer great alterations due to immune response. Then, there is a need for the use of multivariate analysis through machine learning algorithms, which allow us to access the main differences and classify the samples according to their groups. Recently, researchers demonstrated the potential use of FTIR spectroscopy associated with machine learning algorithms for canine VL diagnosis [20]. We showed that alterations in molecular vibrational modes, assigned to the amides I and II group, have a great contribution in multivariate analysis for group classification, making it possible to generate a diagnostic method with an overall accuracy above 85%. Here, we aimed to develop an alternative diagnostic method based on UV spectroscopy associated with machine learning for VL diagnosis by using the antigen-antibody interaction in canine blood serum.