Huan Souza
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Huan Souza has a total of 9 co-authors. Here are the top co-authors by number of shared publications:
- A. C. Lehum
Universidade federal do Pará
14 shared publications - L. Ibiapina Bevilaqua
Escola de Ciências e Tecnologia Universidade Federal do Rio Grande do Norte Caixa Postal 1524 -970 59072 Natal Rio Grande do Norte Brazil
10 shared publications - Pankaj Mehta
Department of Physics, Boston University
3 shared publications - Adilson Jose da Silva
Instituto de Física Universidade de São Paulo Caixa Postal 66318 05315-970 São Paulo SP Brazil
2 shared publications - Albert Yu. Petrov
Universidade Federal da Paraíba
2 shared publications - C. R. Senise Jr.
Univ Fed Sao Paulo, Dept Fis, Rua Sao Nicolau 210, BR-09913030 Diadema, SP, Brazil
2 shared publications
Latest Publications
- Open Access
- 18 February 2026
Parameter-free representations outperform single-cell foundation models on downstream benchmarks
Single-cell RNA sequencing (scRNA-seq) data exhibit strong and reproducible statistical structure. This has motivated the development of large-scale foundation models, such as TranscriptFormer, that use transformer-based architectures to learn a generative model for gene expression by embedding genes into a latent vector space. These embeddings have been used to obtain state-of-the-art (SOTA) performance on downstream tasks such as cell-type classification, disease-state prediction, and cross-species learning. Here, we ask whether similar performance can be achieved without utilizing computationally intensive deep learning-based representations. Using simple, interpretable pipelines that rely on careful normalization and linear methods, we obtain SOTA or near SOTA performance across multiple benchmarks commonly used to evaluate single-cell foundation models, including outperforming foundation models on out-of-distribution tasks involving novel cell types and organisms absent from the training data. Our findings highlight the need for rigorous benchmarking and suggest that the biology of cell identity can be captured by simple linear representations of single cell gene expression data.
- 18 February 2026
Open AccessParameter-free representations outperform single-cell foundation models on downstream benchmarks
Single-cell RNA sequencing (scRNA-seq) data exhibit strong and reproducible statistical structure. This has motivated the development of large-scale foundation models, such as TranscriptFormer, that use transformer-based architectures to learn a generative model for gene expression by embedding genes into a latent vector space. These embeddings have been used to obtain state-of-the-art (SOTA) performance on downstream tasks such as cell-type classification, disease-state prediction, and cross-species learning. Here, we ask whether similar performance can be achieved without utilizing computationally intensive deep learning-based representations. Using simple, interpretable pipelines that rely on careful normalization and linear methods, we obtain SOTA or near SOTA performance across multiple benchmarks commonly used to evaluate single-cell foundation models, including outperforming foundation models on out-of-distribution tasks involving novel cell types and organisms absent from the training data. Our findings highlight the need for rigorous benchmarking and suggest that the biology of cell identity can be captured by simple linear representations of single cell gene expression data.
- Open Access
Parameter-free representations outperform single-cell foundation models on downstream benchmarks
Single-cell RNA sequencing (scRNA-seq) data exhibit strong and reproducible statistical structure. This has motivated the development of large-scale foundation models, such as TranscriptFormer, that use transformer-based architectures to learn a generative model for gene expression by embedding genes into a latent vector space. These embeddings have been used to obtain state-of-the-art (SOTA) performance on downstream tasks such as cell-type classification, disease-state prediction, and cross-species learning. Here, we ask whether similar performance can be achieved without utilizing computationally intensive deep learning-based representations. Using simple, interpretable pipelines that rely on careful normalization and linear methods, we obtain SOTA or near SOTA performance across multiple benchmarks commonly used to evaluate single-cell foundation models, including outperforming foundation models on out-of-distribution tasks involving novel cell types and organisms absent from the training data. Our findings highlight the need for rigorous benchmarking and suggest that the biology of cell identity can be captured by simple linear representations of single cell gene expression data.
Open AccessParameter-free representations outperform single-cell foundation models on downstream benchmarks
Single-cell RNA sequencing (scRNA-seq) data exhibit strong and reproducible statistical structure. This has motivated the development of large-scale foundation models, such as TranscriptFormer, that use transformer-based architectures to learn a generative model for gene expression by embedding genes into a latent vector space. These embeddings have been used to obtain state-of-the-art (SOTA) performance on downstream tasks such as cell-type classification, disease-state prediction, and cross-species learning. Here, we ask whether similar performance can be achieved without utilizing computationally intensive deep learning-based representations. Using simple, interpretable pipelines that rely on careful normalization and linear methods, we obtain SOTA or near SOTA performance across multiple benchmarks commonly used to evaluate single-cell foundation models, including outperforming foundation models on out-of-distribution tasks involving novel cell types and organisms absent from the training data. Our findings highlight the need for rigorous benchmarking and suggest that the biology of cell identity can be captured by simple linear representations of single cell gene expression data.
- Open Access
Simple models outperform single-cell foundation models on downstream benchmarks
Single-cell RNA sequencing (scRNA-seq) data exhibit strong and reproducible statistical structure. This has motivated the development of large-scale foundation models, such as TranscriptFormer, that use transformer-based architectures to learn a generative model for gene expression by embedding genes into a latent vector space. These embeddings have been used to obtain state-of-the-art (SOTA) performance on downstream tasks such as cell-type classification, disease-state prediction, and cross-species learning. Here, we ask whether similar performance can be achieved without utilizing computationally intensive deep learning-based representations. Using simple, interpretable pipelines that rely on careful normalization and linear methods, we obtain SOTA or near SOTA performance across multiple benchmarks commonly used to evaluate single-cell foundation models, including outperforming foundation models on out-of-distribution tasks involving novel cell types and organisms absent from the training data. Our findings highlight the need for rigorous benchmarking and suggest that the biology of cell identity can be captured by simple linear representations of single cell gene expression data.
Open AccessSimple models outperform single-cell foundation models on downstream benchmarks
Single-cell RNA sequencing (scRNA-seq) data exhibit strong and reproducible statistical structure. This has motivated the development of large-scale foundation models, such as TranscriptFormer, that use transformer-based architectures to learn a generative model for gene expression by embedding genes into a latent vector space. These embeddings have been used to obtain state-of-the-art (SOTA) performance on downstream tasks such as cell-type classification, disease-state prediction, and cross-species learning. Here, we ask whether similar performance can be achieved without utilizing computationally intensive deep learning-based representations. Using simple, interpretable pipelines that rely on careful normalization and linear methods, we obtain SOTA or near SOTA performance across multiple benchmarks commonly used to evaluate single-cell foundation models, including outperforming foundation models on out-of-distribution tasks involving novel cell types and organisms absent from the training data. Our findings highlight the need for rigorous benchmarking and suggest that the biology of cell identity can be captured by simple linear representations of single cell gene expression data.
- Open Access
- 13 June 2023
- Published by American Physical Society (APS) in Physical Review D
- Vol. 107 (12) , 125012
Gravitational corrections to the Einstein-scalar-QCD model
This study employs the effective field theory approach to quantum gravity to investigate a non-Abelian gauge theory involving scalar particles coupled to gravity. The study demonstrates explicitly that the Slavnov-Taylor identities are maintained at one-loop order, which indicates that the universality of the color charge is preserved. Additionally, the graviton corrections to the two-loop gluon self-energy and its renormalization are computed.
- 13 June 2023
- Published by American Physical Society (APS) in Physical Review D
- Vol. 107 (12) , 125012
Open AccessGravitational corrections to the Einstein-scalar-QCD model
This study employs the effective field theory approach to quantum gravity to investigate a non-Abelian gauge theory involving scalar particles coupled to gravity. The study demonstrates explicitly that the Slavnov-Taylor identities are maintained at one-loop order, which indicates that the universality of the color charge is preserved. Additionally, the graviton corrections to the two-loop gluon self-energy and its renormalization are computed.
- Open Access
Gravitational corrections to the Einstein-Scalar-QCD model
This study employs the effective field theory approach to quantum gravity to investigate a non-Abelian gauge theory involving scalar particles coupled to gravity. The study demonstrates explicitly that the Slavnov-Taylor identities are maintained at one-loop order, which indicates that the universality of the color charge is preserved. Additionally, the graviton corrections to the two-loop gluon self-energy and its renormalization are computed.
Open AccessGravitational corrections to the Einstein-Scalar-QCD model
This study employs the effective field theory approach to quantum gravity to investigate a non-Abelian gauge theory involving scalar particles coupled to gravity. The study demonstrates explicitly that the Slavnov-Taylor identities are maintained at one-loop order, which indicates that the universality of the color charge is preserved. Additionally, the graviton corrections to the two-loop gluon self-energy and its renormalization are computed.
- Open Access
Improved effective potential of Gildener-Weinberg models
The Gildener-Weinberg models are of particular interest in the context of extensions to the Standard Model of particle physics. These extensions may encompass a variety of theories, including double Higgs models, Grand Unification Theories, and proposals for Dark Matter, among others. In order to rigorously test these models experimentally, obtaining precise results is of crucial importance. In this study, we employ the renormalization group equation and its one-loop functions to obtain a deeper understanding of the higher-loop effective potential. Our findings reveal that the radiatively generated mass of the light particle in the Gildener-Weinberg approach experiences a substantial correction. Furthermore, our results suggest that not all flat directions are equivalent and some may be preferred by nature.
Open AccessImproved effective potential of Gildener-Weinberg models
The Gildener-Weinberg models are of particular interest in the context of extensions to the Standard Model of particle physics. These extensions may encompass a variety of theories, including double Higgs models, Grand Unification Theories, and proposals for Dark Matter, among others. In order to rigorously test these models experimentally, obtaining precise results is of crucial importance. In this study, we employ the renormalization group equation and its one-loop functions to obtain a deeper understanding of the higher-loop effective potential. Our findings reveal that the radiatively generated mass of the light particle in the Gildener-Weinberg approach experiences a substantial correction. Furthermore, our results suggest that not all flat directions are equivalent and some may be preferred by nature.
- Open Access
- 8 May 2023
- Published by American Physical Society (APS) in Physical Review D
- Vol. 107 (10) , 105003
Improved effective potential of Gildener-Weinberg models
The Gildener-Weinberg models are of particular interest in the context of extensions to the Standard Model of particle physics. These extensions may encompass a variety of theories, including double Higgs models, grand unification theories, and proposals for dark matter, among others. In order to rigorously test these models experimentally, obtaining precise results is of crucial importance. In this study, we employ the renormalization group equation and its one-loop functions to obtain a deeper understanding of the higher-loop effective potential. Our findings reveal that the radiatively generated mass of the light particle in the Gildener-Weinberg approach experiences a substantial correction. Furthermore, our results suggest that not all flat directions are equivalent and some may be preferred by nature.
- 8 May 2023
- Published by American Physical Society (APS) in Physical Review D
- Vol. 107 (10) , 105003
Open AccessImproved effective potential of Gildener-Weinberg models
The Gildener-Weinberg models are of particular interest in the context of extensions to the Standard Model of particle physics. These extensions may encompass a variety of theories, including double Higgs models, grand unification theories, and proposals for dark matter, among others. In order to rigorously test these models experimentally, obtaining precise results is of crucial importance. In this study, we employ the renormalization group equation and its one-loop functions to obtain a deeper understanding of the higher-loop effective potential. Our findings reveal that the radiatively generated mass of the light particle in the Gildener-Weinberg approach experiences a substantial correction. Furthermore, our results suggest that not all flat directions are equivalent and some may be preferred by nature.
- Open Access
- 10 August 2022
- Published by American Physical Society (APS) in Physical Review D
- Vol. 106 (4) , 045010
Gravitational corrections to a non-Abelian gauge theory
This paper is part of a series of papers exploring the renormalization of field theories coupled to gravity using the effective field theory framework. In previous works, we studied the universality of the electric charge and the two-loops beta function in the Einstein-QED system. Now, we use this framework to study the non-Abelian gauge theory with fermions coupled to gravity. We show that even though some of the counterterms are dependent on the gravitational coupling, the renormalized coupling constant, and therefore the beta function, do not receive any gravitational correction at one-loop order. Also, we explicitly show that, at this order, the Slavnov-Taylor identities are respected. Finally, we end the paper by putting into perspective the results of the present and the past papers.
- 10 August 2022
- Published by American Physical Society (APS) in Physical Review D
- Vol. 106 (4) , 045010
Open AccessGravitational corrections to a non-Abelian gauge theory
This paper is part of a series of papers exploring the renormalization of field theories coupled to gravity using the effective field theory framework. In previous works, we studied the universality of the electric charge and the two-loops beta function in the Einstein-QED system. Now, we use this framework to study the non-Abelian gauge theory with fermions coupled to gravity. We show that even though some of the counterterms are dependent on the gravitational coupling, the renormalized coupling constant, and therefore the beta function, do not receive any gravitational correction at one-loop order. Also, we explicitly show that, at this order, the Slavnov-Taylor identities are respected. Finally, we end the paper by putting into perspective the results of the present and the past papers.
- Open Access
Gravitational corrections to a non-Abelian gauge theory
This paper is part of a series of papers exploring the renormalization of field theories coupled to gravity using the effective field theory framework. In previous works we studied the universality of the electric charge and the two-loops beta function in the Einstein-QED system. Now, we use this framework to study the non-Abelian gauge theory with fermions coupled to gravity. We show that even though some of the counterterms are dependent on the gravitational coupling, the renormalized coupling constant, and therefore the beta function, do not receive any gravitational correction at one-loop order. Also, we explicitly show that, at this order, the Slavnov-Taylor identities are respected. Finally, we end the paper putting into perspective the results of the present and the past papers.
Open AccessGravitational corrections to a non-Abelian gauge theory
This paper is part of a series of papers exploring the renormalization of field theories coupled to gravity using the effective field theory framework. In previous works we studied the universality of the electric charge and the two-loops beta function in the Einstein-QED system. Now, we use this framework to study the non-Abelian gauge theory with fermions coupled to gravity. We show that even though some of the counterterms are dependent on the gravitational coupling, the renormalized coupling constant, and therefore the beta function, do not receive any gravitational correction at one-loop order. Also, we explicitly show that, at this order, the Slavnov-Taylor identities are respected. Finally, we end the paper putting into perspective the results of the present and the past papers.
- Open Access
- 28 December 2021
- Published by American Physical Society (APS) in Physical Review D
- Vol. 104 (12) , 125019
Universality of gauge coupling constant in the Einstein-QED system
In this paper, we discuss the universality of the renormalization of the gauge coupling constant in the quantum electrodynamics coupled to the Einstein’s gravity in the framework of effective field theory in an arbitrary gauge. We observe that the renormalization of the three-point functions with different particles receive different contributions from the gravitational sector. Despite that, the universality of the gauge coupling constants is guaranteed by the Ward identity that makes the renormalized electric charge receive no contribution from the gravitational sector at one-loop order.
- 28 December 2021
- Published by American Physical Society (APS) in Physical Review D
- Vol. 104 (12) , 125019
Open AccessUniversality of gauge coupling constant in the Einstein-QED system
In this paper, we discuss the universality of the renormalization of the gauge coupling constant in the quantum electrodynamics coupled to the Einstein’s gravity in the framework of effective field theory in an arbitrary gauge. We observe that the renormalization of the three-point functions with different particles receive different contributions from the gravitational sector. Despite that, the universality of the gauge coupling constants is guaranteed by the Ward identity that makes the renormalized electric charge receive no contribution from the gravitational sector at one-loop order.