Date of Publication
11-27-2025
Document Type
Master's Thesis
Degree Name
Master of Science in Chemistry
Subject Categories
Chemistry
College
College of Science
Department/Unit
Chemistry
Thesis Advisor
Aldrin P. Bonto
Defense Panel Chair
Raymond S. Malabed
Defense Panel Member
Katherine Yasmin M. Garcia
Vincent Antonio S. Ng
Abstract (English)
Rice (Oryza sativa) is one of the world’s most widely cultivated cereal crops and serves as a staple food for over half of the global population. Pigmented rice varieties, distinguished by their red to black pericarp, are increasingly regarded as nutritionally superior to non-pigmented rice due to their high levels of health-promoting phenolic compounds concentrated in the bran. Traditional spectrophotometric and chromatographic assays for quantifying key phytochemicals, total phenolic content (TPC), total flavonoid content (TFC), total anthocyanin content (TAC), and total proanthocyanidin content (TPAC), are accurate but labor-intensive, requiring extensive sample preparation, derivatization, and careful optimization of analytical parameters. Attenuated Total Reflectance–Fourier Transform Infrared (ATR-FTIR) spectroscopy offers a rapid, minimally preparative alternative for assessing these compounds. In this study, TPC, TFC, TAC, and TPAC were quantified in 239 pigmented rice varieties, and mid-IR spectra of their extracts were collected within the 4000–400 cm⁻¹ region. Several regression algorithms—k-Nearest Neighbor (KNN), Decision Tree (DT), Gradient Boosting (GB), and Artificial Neural Network (ANN)—were evaluated to establish predictive models. Prior to modeling, the FTIR spectra were subjected to multiple preprocessing techniques, including minimum–maximum normalization (MMN), vector normalization (VN), and standard normal variate (SNV), each combined with first-derivative (+1D) and second-derivative (+2D) transformations. The optimal model–pretreatment combinations were ANN with VN + 2D for TPC, KNN with VN + 1D for TFC, GB with MMN + 1D for TAC, and GB with VN + 1D for TPAC. The corresponding best prediction accuracies were 34.93% (TPC), 35.63% (TFC), 70.16% (TAC), and 38.71% (TPAC). Although predictive performance varied across phytochemical classes, the results demonstrate the promise of integrating FTIR spectroscopy with machine-learning approaches as a high-throughput, cost-effective strategy for profiling phenolic compounds in pigmented rice with minimal sample preparation.
Abstract Format
html
Abstract (Filipino)
Ang palay (Oryza sativa) ay isa sa mga pinaka-malawak pananim na butil sa buong mundo at nagsisilbing pangunahing pagkain ng mahigit kalahati ng populasyon ng mundo. Ang mga may kulay o pigmented na uri ng bigas, na may pericarp na mula pula hanggang itim ang kulay, ay mas pinahahalagahan sa kasalukuyan dahil mas mayaman ang mga ito sa mga phenolic compound na may mahalagang benepisyo sa kalusugan. Ang mga sangkap na ito ay mas maraming matatagpuan sa ipa kumpara sa mga hindi pigmented na uri ng bigas. Ang mga karaniwang pamamaraang ginagamit sa pagsukat ng mahahalagang phytochemical tulad ng total phenolic content (TPC), total flavonoid content (TFC), total anthocyanin content (TAC), at total proanthocyanidin content (TPAC) ay ang spectrophotometric at chromatographic assay. Gayunpaman, ang mga pamamaraang ito ay matrabaho dahil nangangailangan ng masusing paghahanda ng sample, derivatization, at masusing pag-aayos ng mga parametro. Bilang mas mabilis na alternatibo, nag-aalok ang Attenuated Total Reflectance–Fourier Transform Infrared (ATR-FTIR) spectroscopy ng paraan ng pagsusuri na halos hindi na nangangailangan ng elaboradong paghahanda ng sample. Sa pag-aaral na ito, sinukat ang TPC, TFC, TAC, at TPAC ng 239 na uri ng pigmented na bigas. Kinuha rin ang kanilang mid-infrared (mid-IR) spectra sa saklaw na 4000–400 cm⁻¹. Gumamit ang mga mananaliksik ng iba’t ibang regression algorithm tulad ng k-Nearest Neighbor (KNN), Decision Tree (DT), Gradient Boosting (GB), at Artificial Neural Network (ANN) upang makabuo ng mga modelong ginagamit sa prediksyon. Bago isagawa ang pagmomodelo, isinailalim muna ang FTIR spectra sa iba’t ibang preprocessing techniques gaya ng minimum–maximum normalization (MMN), vector normalization (VN), at standard normal variate (SNV), na sinabayan ng first-derivative (+1D) at second-derivative (+2D) na mga transpormasyon. Napag-alaman na ang pinakamainam na kombinasyon ng modelo at pretreatment ay ang mga sumusunod: ANN na may VN + 2D para sa TPC, KNN na may VN + 1D para sa TFC, GB na may MMN + 1D para sa TAC, at GB na may VN + 1D para sa TPAC. Ang pinakamataas na naitalang prediction accuracy ay 34.93% para sa TPC, 35.63% para sa TFC, 70.16% para sa TAC, at 38.71% para sa TPAC. Bagama’t magkakaiba ang antas ng pagiging tumpak ng prediksyon para sa bawat uri ng phytochemical, malinaw na ipinakikita ng mga resulta na may malaking potensyal ang pagsasama ng FTIR spectroscopy at mga machine-learning approach bilang isang mabilis, matipid, at epektibong paraan para sa pagsusuri at pag-profile ng mga phenolic compound sa pigmented na bigas, na nangangailangan lamang ng kaunting paghahanda ng sample.
Abstract Format
html
Language
English
Format
Electronic
Keywords
Rice; Fourier transform infrared spectroscopy Machine learning; Chemometrics
Recommended Citation
Seagan, C. V. (2025). Prediction of polyphenolic content of pigmented rice bran extracts using ATR-mid-FTIR spectroscopy in tandem with multivariate machine learning models. Retrieved from https://animorepository.dlsu.edu.ph/etdm_chem/32
Upload Full Text
wf_yes
Embargo Period
12-9-2028
