Machine learning-guided infrared spectroscopy identifies distinct biochemical shifts tied to Parkinson’s disease in fruit flies, offering a rapid, label-free diagnostic framework.
Researchers have found a novel way to spot the biochemical signatures of Parkinson’s disease in fruit flies using a combination of advanced laser-based light scanning and artificial intelligence. Parkinson’s disease is the second most common neurodegenerative disorder globally, trailing only Alzheimer’s disease. It is clinically characterised by the progressive death of movement-controlling dopamine-producing brain cells, alongside high levels of cellular oxidative stress, mitochondrial failure, and metabolic breakdown.
Conventionally, tracking these molecular shifts requires complex, targeted chemical tests or assays that can be slow and expensive. A team of researchers at the CSIR-Central Institute of Medicinal and Aromatic Plants (CSIR-CMAP) has now shown that a technique known as attenuated total reflectance Fourier transform infrared spectroscopy can map an organism’s entire molecular fingerprint at once, providing a fast, non-destructive, label-free window into disease progression.
To test whether infrared light could reliably capture Parkinson’s pathology, the researchers turned to the common fruit fly, Drosophila melanogaster. The organism is widely used in laboratory settings because of its well-mapped neurobehavioral and genetic traits. The team exposed groups of male flies to three well-established chemical neurotoxicants known to mimic Parkinson’s disease symptoms: paraquat, rotenone, and MPTP.
Before scanning the insects, the researchers validated that the toxin-exposed flies truly developed Parkinson’s-like molecular changes or phenotypes. Behavioural tests, such as negative geotaxis climbing assays and continuous circadian locomotor monitoring, showed that the treated flies suffered severe motor coordination problems, sleep-wake rhythm disruptions, and generalised sluggishness. Furthermore, microscopic brain examinations using specific antibody stains confirmed the loss of crucial dopaminergic neurons, structural tissue decay, a collapse in mitochondrial membrane potential, and a spike in damaging reactive oxygen species and fat droplet accumulation.
Once the researchers validated the disease models, they dissected intact fly heads and analysed them using infrared spectroscopy across the mid-infrared region. Every biological molecule vibrates at specific frequencies when hit with infrared light, creating a unique absorption spectrum that acts like a molecular barcode. In their tests, the raw spectral data revealed distinct alterations in regions associated with lipids, proteins, and cellular fingerprints, reflecting major molecular remodelling caused by neurodegeneration.
Because these spectral datasets are massive and full of overlapping signals, the team applied machine learning algorithms, including partial least squares discriminant analysis, support vector machines, and random forest models. By filtering the data through different variables, the algorithms successfully isolated key spectral changes that allowed computers to accurately separate healthy control flies from the Parkinson’s disease models with impressive classification accuracy.
This research successfully merges classical chemometrics, modern machine learning classifiers, and rigorous one-class modelling into a single, unified framework. By pairing spectral analysis with direct biological tracking of reactive oxygen species and lipid remodelling, the team showed that their computational models were capturing genuine neurodegenerative distress rather than random statistical noise. Furthermore, the use of independent test datasets that remained entirely hidden during model training ensures strong generalisation, showing the technology is robust and reliable.
Despite its potential, the authors note that the infrared technique only captures a generalised biochemical snapshot rather than pointing directly to individual molecular species. Future research will need to expand toward multi-class classification strategies, larger independent datasets, and complementary molecular checks to refine the technology further.
Neurodegenerative disorders often cause severe cellular damage long before outward clinical symptoms ever appear, creating a desperate need for early diagnostic tools. By proving that infrared spectroscopy coupled with artificial intelligence can rapidly and accurately identify the biochemical fingerprints of neurodegeneration, this study paves the way for future high-throughput, label-free screening methods. Such innovations could eventually lower diagnostic barriers and detect neurological decline at its earliest, most manageable stages.
inson’s disease-associated biochemical changes in Drosophila melanogaster
Machine-learningParkinson’sAcademy of Scientific and Innovative Researchparkinson diseaseCSIR-Central Institute of Medicinal and Aromatic Plants (CSIR-CMAP)
