Rogerio Salvini
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Rogerio Salvini has a total of 80 co-authors. Here are the top co-authors by number of shared publications:
- Fabrizzio Soares
Universidade Federal de Goiás
7 shared publications - Juliana Paula Felix
Universidade Federal de Goiás
5 shared publications - Giordana de Farias F. B. Bucci
Universidade Federal de Goiás
3 shared publications - Ilaria Gabbatore
Department of Humanities, University of Turin, Turin, Italy
3 shared publications Pedro L. S. Lobo
Instituto de Informática Universidade Federal de Goiás
3 shared publications- Afonso Ueslei Fonseca
Universidade Federal de Goiás
2 shared publications
Latest Publications
- Open Access
- 3 April 2026
- Published by Elsevier in SSRN Electronic Journal
A comprehensive study to compare different compound representations for predicting carcinogenicity in vivo
Carcinogenicity evaluation is a critical component of chemical risk assessment, yet traditional in vivo testing remains time-consuming, costly, and ethically challenging. Computational approaches based on machine learning offer promising alternatives, but the relative contribution of different molecular representation strategies for predicting in vivo carcinogenicity remains insufficiently explored. This study aimed to systematically evaluate the impact of molecular embeddings, classical descriptors, and toxicophore structural alerts on the performance of machine learning models for in vivo carcinogenicity prediction. A curated dataset of 2,090 distinct compounds tested exclusively in vivo was assembled by integrating five major toxicological databases. Compounds were represented using classical molecular descriptors, descriptor sets enriched with structural alerts, SMILES-derived molecular embeddings, and hybrid combinations of these representations. Twenty-four machine learning classifiers were benchmarked under a 10-fold stratified cross-validation protocol. Model performance was assessed using accuracy, precision, recall, F1-score, and AUC-ROC, with statistical significance evaluated using Friedman and Nemenyi tests. Results indicated that representations combining molecular descriptors with structural alerts tend to yield the most consistent predictive performance across models. Embeddings contribute as complementary features but do not replace classical representations. These findings reinforce the central role of chemically interpretable, expert-driven descriptors, particularly those incorporating genotoxic structural alerts, in regulatory-relevant carcinogenicity modeling.
- 3 April 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessA comprehensive study to compare different compound representations for predicting carcinogenicity in vivo
Carcinogenicity evaluation is a critical component of chemical risk assessment, yet traditional in vivo testing remains time-consuming, costly, and ethically challenging. Computational approaches based on machine learning offer promising alternatives, but the relative contribution of different molecular representation strategies for predicting in vivo carcinogenicity remains insufficiently explored. This study aimed to systematically evaluate the impact of molecular embeddings, classical descriptors, and toxicophore structural alerts on the performance of machine learning models for in vivo carcinogenicity prediction. A curated dataset of 2,090 distinct compounds tested exclusively in vivo was assembled by integrating five major toxicological databases. Compounds were represented using classical molecular descriptors, descriptor sets enriched with structural alerts, SMILES-derived molecular embeddings, and hybrid combinations of these representations. Twenty-four machine learning classifiers were benchmarked under a 10-fold stratified cross-validation protocol. Model performance was assessed using accuracy, precision, recall, F1-score, and AUC-ROC, with statistical significance evaluated using Friedman and Nemenyi tests. Results indicated that representations combining molecular descriptors with structural alerts tend to yield the most consistent predictive performance across models. Embeddings contribute as complementary features but do not replace classical representations. These findings reinforce the central role of chemically interpretable, expert-driven descriptors, particularly those incorporating genotoxic structural alerts, in regulatory-relevant carcinogenicity modeling.
- Open Access
- 29 December 2025
- Published by Sociedade Brasileira de Computação in Revista Eletrônica de Iniciação Científica em Computação
- Vol. 23, 291-297
Investigação de Enviesamento de Modelos de Aprendizado de Máquina para Diagnóstico de Doenças Neurodegenerativas via Registros de Marcha e Voz
Este trabalho sintetiza os principais achados de uma pesquisa sobre vieses em modelos de aprendizado de máquina para o diagnóstico de doenças neurodegenerativas por análise da marcha e da voz. Avaliamos como técnicas de sobreamostragem, como o janelamento da marcha e o uso indiscriminado de múltiplas amostras de voz por pessoa, inflam métricas de desempenho ao tratar amostras do mesmo sujeito como independentes. Comparando protocolos que ignoram ou preservam essas dependências em dois conjuntos de dados, observamos que desconsiderá-las inflaciona as métricas, enquanto sua preservação fornece avaliações mais fiáveis. Os resultados destacam a importância da segregação adequada das amostras para obter modelos diagnósticos confiáveis.
- 29 December 2025
- Published by Sociedade Brasileira de Computação in Revista Eletrônica de Iniciação Científica em Computação
- Vol. 23, 291-297
Open AccessInvestigação de Enviesamento de Modelos de Aprendizado de Máquina para Diagnóstico de Doenças Neurodegenerativas via Registros de Marcha e Voz
Este trabalho sintetiza os principais achados de uma pesquisa sobre vieses em modelos de aprendizado de máquina para o diagnóstico de doenças neurodegenerativas por análise da marcha e da voz. Avaliamos como técnicas de sobreamostragem, como o janelamento da marcha e o uso indiscriminado de múltiplas amostras de voz por pessoa, inflam métricas de desempenho ao tratar amostras do mesmo sujeito como independentes. Comparando protocolos que ignoram ou preservam essas dependências em dois conjuntos de dados, observamos que desconsiderá-las inflaciona as métricas, enquanto sua preservação fornece avaliações mais fiáveis. Os resultados destacam a importância da segregação adequada das amostras para obter modelos diagnósticos confiáveis.
- 4 December 2025
- Published by Sociedade Brasileira de Computação
- p. 344-347
Parkinson’s Disease Classification Using Inertial Signals from Smartwatches and Tree-Based Classifiers
This study evaluates the use of statistical features extracted from inertial signals obtained from smartwatches to assist in distinguishing patients with Parkinson’s disease, differential diagnoses, and healthy controls. Using a public dataset comprising data from 469 individuals, tree-based classifiers were applied. The results indicate that simple motion information can support the diagnosis of Parkinson’s disease in a non-invasive and accessible manner.
- 4 December 2025
- Published by Sociedade Brasileira de Computação
- p. 344-347
Parkinson’s Disease Classification Using Inertial Signals from Smartwatches and Tree-Based Classifiers
This study evaluates the use of statistical features extracted from inertial signals obtained from smartwatches to assist in distinguishing patients with Parkinson’s disease, differential diagnoses, and healthy controls. Using a public dataset comprising data from 469 individuals, tree-based classifiers were applied. The results indicate that simple motion information can support the diagnosis of Parkinson’s disease in a non-invasive and accessible manner.
- 4 December 2025
- Published by Sociedade Brasileira de Computação
- p. 348-351
Robustness and Cross-Dataset Performance of Machine Learning Models in Parkinson’s Disease Diagnosis
Parkinson’s disease (PD) is a progressive neurodegenerative disorder whose diagnosis remains largely based on subjective clinical evaluation. In this study, we examine the generalization capacity of machine learning models for PD diagnosis from gait data. Descriptive features extracted from force signals were used to train multiple classifiers, evaluated through intra-dataset and cross-dataset experiments. Results showed consistent performance, particularly for ensemble methods such as ExtraTrees, demonstrating the robustness of the proposed approach. These findings highlight the importance of cross-dataset validation for assessing the real-world applicability of diagnostic models.
- 4 December 2025
- Published by Sociedade Brasileira de Computação
- p. 348-351
Robustness and Cross-Dataset Performance of Machine Learning Models in Parkinson’s Disease Diagnosis
Parkinson’s disease (PD) is a progressive neurodegenerative disorder whose diagnosis remains largely based on subjective clinical evaluation. In this study, we examine the generalization capacity of machine learning models for PD diagnosis from gait data. Descriptive features extracted from force signals were used to train multiple classifiers, evaluated through intra-dataset and cross-dataset experiments. Results showed consistent performance, particularly for ensemble methods such as ExtraTrees, demonstrating the robustness of the proposed approach. These findings highlight the importance of cross-dataset validation for assessing the real-world applicability of diagnostic models.
- 4 December 2025
- Published by Sociedade Brasileira de Computação
- p. 175-184
Identificação da Doença de Parkinson através da Análise de Dados Coletados por Smartwatch
A doença de Parkinson (DP) é uma doença neurodegenerativa de difícil diagnóstico precoce que causa degeneração progressiva dos neurônios, provocando alterações na marcha e movimentos. Métodos diagnósticos computacionais baseados em aprendizado de máquina são uma possível solução para essa problemática. Este trabalho reporta os principais resultados obtidos investigando o uso de características harmônicas extraídas de sinais de movimento e de algorítimos clássicos de aprendizado de máquina supervisionado para a classificação da Doença de Parkinson. Os resultados evidenciam o potencial de algoritmos de aprendizado de máquina para auxiliar nesse propósito.
- 4 December 2025
- Published by Sociedade Brasileira de Computação
- p. 175-184
Identificação da Doença de Parkinson através da Análise de Dados Coletados por Smartwatch
A doença de Parkinson (DP) é uma doença neurodegenerativa de difícil diagnóstico precoce que causa degeneração progressiva dos neurônios, provocando alterações na marcha e movimentos. Métodos diagnósticos computacionais baseados em aprendizado de máquina são uma possível solução para essa problemática. Este trabalho reporta os principais resultados obtidos investigando o uso de características harmônicas extraídas de sinais de movimento e de algorítimos clássicos de aprendizado de máquina supervisionado para a classificação da Doença de Parkinson. Os resultados evidenciam o potencial de algoritmos de aprendizado de máquina para auxiliar nesse propósito.
- Open Access
- 5 December 2024
- Published by Sociedade Brasileira de Computação
- p. 158-167
Aplicação de Algoritmos de Aprendizado de Máquina na Análise da Vulnerabilidade Social e Insegurança Alimentar
A fome é um problema político e econômico com profundas repercussões sociais, sendo a vulnerabilidade social um indicador essencial para essa compreensão. Este estudo aplica métodos de aprendizado de máquina para analisar a insegurança alimentar no contexto da vulnerabilidade social. Dados socioeconômicos da Fundação SEADE, RAIS e CAISAN foram usados para gerar modelos de classificação que alcançaram F-score de 80% a 87% na categorização do Índice Paulista de Vulnerabilidade Social (IPVS). A renda domiciliar destacou-se como o fator mais relevante. Os resultados corroboram estudos anteriores, apontando que dados socioeconômicos podem ser explorados na identificação de indicadores de vulnerabilidade e insegurança alimentar.
- 5 December 2024
- Published by Sociedade Brasileira de Computação
- p. 158-167
Open AccessAplicação de Algoritmos de Aprendizado de Máquina na Análise da Vulnerabilidade Social e Insegurança Alimentar
A fome é um problema político e econômico com profundas repercussões sociais, sendo a vulnerabilidade social um indicador essencial para essa compreensão. Este estudo aplica métodos de aprendizado de máquina para analisar a insegurança alimentar no contexto da vulnerabilidade social. Dados socioeconômicos da Fundação SEADE, RAIS e CAISAN foram usados para gerar modelos de classificação que alcançaram F-score de 80% a 87% na categorização do Índice Paulista de Vulnerabilidade Social (IPVS). A renda domiciliar destacou-se como o fator mais relevante. Os resultados corroboram estudos anteriores, apontando que dados socioeconômicos podem ser explorados na identificação de indicadores de vulnerabilidade e insegurança alimentar.
- Open Access
- 18 May 2022
- Published by Universidade de São Paulo. Agência de Bibliotecas e Coleções Digitais in Revista de Saúde Pública
- Vol. 56, 38
Microcephaly measurement in adults and its association with clinical variables
OBJECTIVE: To establish a microcephaly cut-off size in adults using head circumference as an indirect measure of brain size, as well as to explore factors associated with microcephaly via data mining. METHODS: In autopsy studies, head circumference was measured with an inelastic tape placed around the skull. Total brain volume was also directly measured. A linear regression was used to determine the association of head circumference with brain volume and clinical variables. Microcephaly was defined as head circumference that were two standard deviations below the mean of significant clinical variables. We further applied an association rule mining to find rules associating microcephaly with several sociodemographic and clinical variables. RESULTS: In our sample of 2,508 adults, the mean head circumference was 55.3 ± 2.7cm. Head circumference was related to height, cerebral volume, and sex (p < 0.001 for all). Microcephaly was present in 4.7% of the sample (n = 119). Out of 34,355 association rules, we found significant relationships between microcephaly and a clinical dementia rating (CDR) > 0.5 with an informant questionnaire on cognitive decline in the elderly (IQCODE) ≥ 3.4 (confidence: 100% and lift: 5.6), between microcephaly and a CDR > 0.5 with age over 70 years (confidence: 42% and lift: 2.4), and microcephaly and males (confidence: 68.1% and lift: 1.3). CONCLUSION: Head circumference was related to cerebral volume. Due to its low cost and easy use, head circumference can be used as a screening test for microcephaly, adjusting it for gender and height. Microcephaly was associated with dementia at old age.
- 18 May 2022
- Published by Universidade de São Paulo. Agência de Bibliotecas e Coleções Digitais in Revista de Saúde Pública
- Vol. 56, 38
Open AccessMicrocephaly measurement in adults and its association with clinical variables
OBJECTIVE: To establish a microcephaly cut-off size in adults using head circumference as an indirect measure of brain size, as well as to explore factors associated with microcephaly via data mining. METHODS: In autopsy studies, head circumference was measured with an inelastic tape placed around the skull. Total brain volume was also directly measured. A linear regression was used to determine the association of head circumference with brain volume and clinical variables. Microcephaly was defined as head circumference that were two standard deviations below the mean of significant clinical variables. We further applied an association rule mining to find rules associating microcephaly with several sociodemographic and clinical variables. RESULTS: In our sample of 2,508 adults, the mean head circumference was 55.3 ± 2.7cm. Head circumference was related to height, cerebral volume, and sex (p < 0.001 for all). Microcephaly was present in 4.7% of the sample (n = 119). Out of 34,355 association rules, we found significant relationships between microcephaly and a clinical dementia rating (CDR) > 0.5 with an informant questionnaire on cognitive decline in the elderly (IQCODE) ≥ 3.4 (confidence: 100% and lift: 5.6), between microcephaly and a CDR > 0.5 with age over 70 years (confidence: 42% and lift: 2.4), and microcephaly and males (confidence: 68.1% and lift: 1.3). CONCLUSION: Head circumference was related to cerebral volume. Due to its low cost and easy use, head circumference can be used as a screening test for microcephaly, adjusting it for gender and height. Microcephaly was associated with dementia at old age.
- Open Access
- 24 May 2021
- Published by Springer Nature in Schizophrenia
- Vol. 7 (1) , 1-9
Multimodal assessment of communicative-pragmatic features in schizophrenia: a machine learning approach
An impairment in pragmatic communication is a core feature of schizophrenia, often associated with difficulties in social interactions. The pragmatic deficits regard various pragmatic phenomena, e.g., direct and indirect communicative acts, deceit, irony, and include not only the use of language but also other expressive means such as non-verbal/extralinguistic modalities, e.g., gestures and body movements, and paralinguistic cues, e.g., prosody and tone of voice. The present paper focuses on the identification of those pragmatic features, i.e., communicative phenomena and expressive modalities, that more reliably discriminate between individuals with schizophrenia and healthy controls. We performed a multimodal assessment of communicative-pragmatic ability, and applied a machine learning approach, specifically a Decision Tree model, with the aim of identifying the pragmatic features that best separate the data into the two groups, i.e., individuals with schizophrenia and healthy controls, and represent their configuration. The results indicated good overall performance of the Decision Tree model, with mean Accuracy of 82%, Sensitivity of 76%, and Precision of 91%. Linguistic irony emerged as the most relevant pragmatic phenomenon in distinguishing between the two groups, followed by violation of the Gricean maxims, and then extralinguistic deceitful and sincere communicative acts. The results are discussed in light of the pragmatic theoretical literature, and their clinical relevance in terms of content and design of both assessment and rehabilitative training.
- 24 May 2021
- Published by Springer Nature in Schizophrenia
- Vol. 7 (1) , 1-9
Open AccessMultimodal assessment of communicative-pragmatic features in schizophrenia: a machine learning approach
An impairment in pragmatic communication is a core feature of schizophrenia, often associated with difficulties in social interactions. The pragmatic deficits regard various pragmatic phenomena, e.g., direct and indirect communicative acts, deceit, irony, and include not only the use of language but also other expressive means such as non-verbal/extralinguistic modalities, e.g., gestures and body movements, and paralinguistic cues, e.g., prosody and tone of voice. The present paper focuses on the identification of those pragmatic features, i.e., communicative phenomena and expressive modalities, that more reliably discriminate between individuals with schizophrenia and healthy controls. We performed a multimodal assessment of communicative-pragmatic ability, and applied a machine learning approach, specifically a Decision Tree model, with the aim of identifying the pragmatic features that best separate the data into the two groups, i.e., individuals with schizophrenia and healthy controls, and represent their configuration. The results indicated good overall performance of the Decision Tree model, with mean Accuracy of 82%, Sensitivity of 76%, and Precision of 91%. Linguistic irony emerged as the most relevant pragmatic phenomenon in distinguishing between the two groups, followed by violation of the Gricean maxims, and then extralinguistic deceitful and sincere communicative acts. The results are discussed in light of the pragmatic theoretical literature, and their clinical relevance in terms of content and design of both assessment and rehabilitative training.
- Open Access
- 15 December 2020
- Published by Corporacion Universitaria Iberoamericana
Autismo caminhos para a inclusão
Este livro é o resultado do IV Simpósio sobre Ambientes Digitais de Aprendizagem para Crianças Autistas (IV SAACA), evento bianual realizado pela Universidade Federal Fluminense campus Aterrado em Volta Redonda – RJ, que, em sua quarta edição, reuniu profissionais que trabalham com a temática do Autismo e possuem estudos e práticas de relevância na área. Como resultado desse trabalho, nosso interesse em partilhar suas perspectivas, estudos e saberes através deste livro, visa a alcançar um maior número de pessoas que se interessam pelo tema.
- 15 December 2020
- Published by Corporacion Universitaria Iberoamericana
Open AccessAutismo caminhos para a inclusão
Este livro é o resultado do IV Simpósio sobre Ambientes Digitais de Aprendizagem para Crianças Autistas (IV SAACA), evento bianual realizado pela Universidade Federal Fluminense campus Aterrado em Volta Redonda – RJ, que, em sua quarta edição, reuniu profissionais que trabalham com a temática do Autismo e possuem estudos e práticas de relevância na área. Como resultado desse trabalho, nosso interesse em partilhar suas perspectivas, estudos e saberes através deste livro, visa a alcançar um maior número de pessoas que se interessam pelo tema.