L’astrobiology is a discipline that investigates the origin of life on Earth and the possibility of its existence on other planets. A fundamental objective of astrobiological research is the identification of biosignatures, namely chemical, physical, or spectral signals whose biological origin can be distinguished from abiotic processes. Photosynthetic microorganisms, such as microalgae and cyanobacteria, are considered among the first organisms to have colonized Earth. For this reason, the study of biosignatures derived from their photosynthetic activity through Remote Sensing (RS) techniques is of primary importance in astrobiology. Their spectral properties depend on the structure of the photosynthetic apparatus, pigment composition (chlorophylls, carotenoids, and phycobiliproteins), cellular organization, and biomass. Consequently, different species produce potentially species-specific spectral signatures. In particular, cyanobacteria and microalgae, owing to their high phylogenetic and ecological diversity and the presence of characteristic signals in the visible and near-infrared regions (VNIR, 400–1100 nm), represent ideal models for the development of reference spectral libraries. However, their use as taxonomic "barcodes" through machine learning approaches remains largely unexplored. The present study evaluated the possibility of discriminating photosynthetic microorganisms using VNIR reflectance spectra and machine learning algorithms. Thirteen species (7 cyanobacteria and 6 microalgae) were cultivated under standardized conditions, with 10 pseudo-replicates per species. Reflectance was measured using an Ocean Optics Flame-T spectrometer coupled with a 3D-printed optical system designed to standardize the acquisition geometry and incident illumination, thereby reducing technical variability. Reflectance spectra were acquired both in liquid medium and after deposition onto filter paper. Pigment composition was determined by Reverse Phase High-Performance Liquid Chromatography (RP-HPLC). Analyses were performed on both raw reflectance data and data pre-processed using the Savitzky–Golay filter with first derivative, which effectively reduces noise while preserving the shape of the original spectral features, enhances local spectral variations, and improves the reliability of predictive models. Random Forest classification models were developed using either raw or transformed reflectance as predictors. Model validation was performed both internally using a jackknife procedure and externally using an independent dataset containing reflectance spectra from additional cyanobacterial and microalgal species. VNIR reflectance spectra enabled high classification accuracy even when using raw reflectance (OA = 0.97 in liquid medium; OA = 0.78 on filter paper), which was further improved after Savitzky–Golay preprocessing (OA = 1.00 and OA = 0.99, respectively). The VNIR and Red Edge (680–750 nm) regions were identified as the most informative, while RP-HPLC analyses confirmed the close relationship between pigment composition and spectral signatures. External validation yielded a mean accuracy of OA = 0.67, indicating a moderate reduction in model performance while maintaining good predictive transferability across independent datasets. The limited taxonomic coverage represented the main limitation of the study, highlighting the need to expand spectral libraries with a larger number of taxa. Nevertheless, the proposed system represents a relatively inexpensive and standardized solution that could facilitate collaboration among laboratories, accelerating the development of extensive spectral libraries and supporting future applications in environmental monitoring, algal biotechnology, ecological and astrobiological remote sensing, and the development of biosignatures for future missions searching for life beyond Earth.
L’astrobiologia è una disciplina che studia l’origine della vita sulla Terra e la possibilità della sua esistenza su altri pianeti. Un obiettivo fondamentale della ricerca astrobiologica è l’identificazione di biofirme (biosignatures), ossia segnali chimici, fisici o spettrali la cui origine biologica sia distinguibile da processi abiotici. I microrganismi fotosintetici, come microalghe e cianobatteri, sono considerati tra i primi organismi ad aver colonizzato la Terra; per questo motivo, lo studio delle biosignatures derivanti dalla loro attività fotosintetica mediante tecniche di telerilevamento (Remote Sensing, RS) riveste un ruolo di primaria importanza. Le loro proprietà spettrali dipendono dalla struttura dell’apparato fotosintetico, dalla composizione pigmentaria (clorofille, carotenoidi e ficobiliproteine), dall’organizzazione cellulare e dalla biomassa. Di conseguenza, specie differenti producono firme spettrali potenzialmente specie-specifiche. In particolare, cianobatteri e microalghe, grazie alla loro elevata diversità filogenetica ed ecologica e alla presenza di segnali caratteristici nelle regioni del visibile e del vicino infrarosso (VNIR, 400–1100 nm), rappresentano modelli ideali per la costruzione di librerie spettrali di riferimento. Tuttavia, il loro utilizzo come "codici a barre" tassonomici mediante approcci di machine learning è ancora poco esplorato. Il presente studio ha valutato la possibilità di discriminare microorganismi fotosintetici utilizzando spettri di riflettanza VNIR e algoritmi di machine learning. Sono state coltivate in condizioni standardizzate 13 specie (7 cianobatteri e 6 microalghe), con 10 pseudo-repliche per specie. La riflettanza è stata misurata mediante uno spettrometro Ocean Optics Flame-T accoppiato a un sistema ottico stampato in 3D, progettato per standardizzare geometria di acquisizione e illuminazione, riducendo la variabilità tecnica. Gli spettri sono stati acquisiti sia in mezzo liquido sia dopo deposizione su carta da filtro. La composizione pigmentaria è stata determinata mediante RP-HPLC (Reverse Phase High Performance Liquid Chromatography). Le analisi sono state condotte sia sulla riflettanza grezza sia dopo l’applicazione del filtro Savitzky-Golay con derivata prima, che riduce il rumore preservando la forma delle caratteristiche spettrali, enfatizza le variazioni locali e migliora l’affidabilità dei modelli predittivi. Sono stati sviluppati modelli di classificazione Random Forest utilizzando come predittori la riflettanza grezza o trasformata. La validazione è stata eseguita sia internamente, mediante procedura jackknife, sia su un dataset esterno indipendente comprendente altre specie di microalghe e cianobatteri. Gli spettri VNIR hanno consentito un’elevata accuratezza già con la riflettanza grezza (OA = 0,97 in mezzo liquido; OA = 0,78 su carta da filtro), ulteriormente migliorata dopo il pretrattamento Savitzky-Golay (OA = 1,00 e OA = 0,99, rispettivamente). Le regioni VNIR e Red Edge (680–750 nm) sono risultate le più informative, mentre le analisi RP-HPLC hanno confermato la stretta relazione tra composizione pigmentaria e firme spettrali. La validazione esterna ha mostrato un’accuratezza media pari a OA = 0,67, indicando una moderata riduzione delle prestazioni ma una buona capacità di trasferimento del modello a dataset indipendenti. La limitata copertura tassonomica rappresenta il principale limite dello studio, evidenziando la necessità di ampliare le librerie spettrali. Nonostante ciò, il sistema proposto costituisce una soluzione economica e standardizzata che potrebbe favorire la collaborazione tra laboratori, accelerando la costruzione di librerie spettrali estese e supportando future applicazioni nel monitoraggio ambientale, nella biotecnologia algale, nel telerilevamento ecologico e astrobiologico e nella ricerca di biosignatures per future missioni di esplorazione della vita oltre la Terra.
The Light of Life: Spectral Signatures of Photosynthetic Microorganisms
TRESSOLDI, ILARIA
2025/2026
Abstract
L’astrobiology is a discipline that investigates the origin of life on Earth and the possibility of its existence on other planets. A fundamental objective of astrobiological research is the identification of biosignatures, namely chemical, physical, or spectral signals whose biological origin can be distinguished from abiotic processes. Photosynthetic microorganisms, such as microalgae and cyanobacteria, are considered among the first organisms to have colonized Earth. For this reason, the study of biosignatures derived from their photosynthetic activity through Remote Sensing (RS) techniques is of primary importance in astrobiology. Their spectral properties depend on the structure of the photosynthetic apparatus, pigment composition (chlorophylls, carotenoids, and phycobiliproteins), cellular organization, and biomass. Consequently, different species produce potentially species-specific spectral signatures. In particular, cyanobacteria and microalgae, owing to their high phylogenetic and ecological diversity and the presence of characteristic signals in the visible and near-infrared regions (VNIR, 400–1100 nm), represent ideal models for the development of reference spectral libraries. However, their use as taxonomic "barcodes" through machine learning approaches remains largely unexplored. The present study evaluated the possibility of discriminating photosynthetic microorganisms using VNIR reflectance spectra and machine learning algorithms. Thirteen species (7 cyanobacteria and 6 microalgae) were cultivated under standardized conditions, with 10 pseudo-replicates per species. Reflectance was measured using an Ocean Optics Flame-T spectrometer coupled with a 3D-printed optical system designed to standardize the acquisition geometry and incident illumination, thereby reducing technical variability. Reflectance spectra were acquired both in liquid medium and after deposition onto filter paper. Pigment composition was determined by Reverse Phase High-Performance Liquid Chromatography (RP-HPLC). Analyses were performed on both raw reflectance data and data pre-processed using the Savitzky–Golay filter with first derivative, which effectively reduces noise while preserving the shape of the original spectral features, enhances local spectral variations, and improves the reliability of predictive models. Random Forest classification models were developed using either raw or transformed reflectance as predictors. Model validation was performed both internally using a jackknife procedure and externally using an independent dataset containing reflectance spectra from additional cyanobacterial and microalgal species. VNIR reflectance spectra enabled high classification accuracy even when using raw reflectance (OA = 0.97 in liquid medium; OA = 0.78 on filter paper), which was further improved after Savitzky–Golay preprocessing (OA = 1.00 and OA = 0.99, respectively). The VNIR and Red Edge (680–750 nm) regions were identified as the most informative, while RP-HPLC analyses confirmed the close relationship between pigment composition and spectral signatures. External validation yielded a mean accuracy of OA = 0.67, indicating a moderate reduction in model performance while maintaining good predictive transferability across independent datasets. The limited taxonomic coverage represented the main limitation of the study, highlighting the need to expand spectral libraries with a larger number of taxa. Nevertheless, the proposed system represents a relatively inexpensive and standardized solution that could facilitate collaboration among laboratories, accelerating the development of extensive spectral libraries and supporting future applications in environmental monitoring, algal biotechnology, ecological and astrobiological remote sensing, and the development of biosignatures for future missions searching for life beyond Earth.| File | Dimensione | Formato | |
|---|---|---|---|
|
Tressoldi_Ilaria.pdf
Accesso riservato
Dimensione
7.8 MB
Formato
Adobe PDF
|
7.8 MB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/112452