Introduction

Fungi, despite being one of the three major eukaryotic kingdoms alongside animals and plants, remain comparatively underexplored. While nearly 150,000 fungal species have been identified, estimates suggest that several million remain undiscovered.1,2 Fungi play a vital role in sustaining life on Earth and are deeply ingrained in our cultural heritage; however, they remain comparatively neglected and poorly understood.3,4 Compared with bacterial and viral pathogens, fungal pathogens have historically received less attention in studies of human infectious diseases.5 Recent estimates suggest that fungal infections are responsible for approximately 1.6 million deaths annually worldwide.6 Fatal outcomes are particularly associated with allergic reactions, mucosal and superficial infections, and chronic fungal diseases. In individuals with compromised immune systems, severe fungal infections can result in mortality rates of up to 50%. Approximately 600 fungal species are known to cause human disease, including life-threatening systemic infections with high fatality rates.7,8 Examples include Aspergillus fumigatus, Cryptococcus neoformans, Histoplasma capsulatum, and Candida albicans.9,10,11 C. albicans is a major cause of mucosal and systemic infections worldwide.12 The virulence factors (VFs) that determine the ability of pathogens to cause candidiasis have a significant impact on the onset and persistence of infection. The incidence of candidiasis remains substantial, particularly in healthcare settings, despite advances in modern medicine. Even with medical intervention, mortality rates can be high.13,14

A critical factor determining fungal pathogenicity is the presence of VFs, specialized proteins that enable pathogens to colonize host tissues, evade immune defenses, and facilitate infection progression.15 These factors allow infections to persist, proliferate, and disseminate by disrupting host homeostasis and modulating immune responses. Interactions among VFs may also enhance the pathogen’s ability to adapt to the host environment. A deeper understanding of VF functions is important for developing effective antifungal strategies and improving diagnostic approaches.16 Virulence-associated proteins can be broadly categorized into toxins, adhesins, hydrolytic enzymes, and invasion factors that interact with host cells and the immune system to promote infection and support fungal survival and growth within the host. Pathogenic microbes possess adhesins that facilitate their attachment to host cells, helping them avoid clearance from mucosal surfaces. A distinctive feature of infectious microbes is their ability to invade host cells and tissues, which may be facilitated by filamentous growth or active penetration.

Additionally, pathogens can exploit host cells as vehicles or disrupt interepithelial connections to overcome cellular barriers.17 To overcome host defenses, fungi such as C. neoformans can employ both direct invasion and the “Trojan horse” mechanism, enabling them to bypass cellular barriers and disseminate into deeper tissues.18,19 Proteases and lipases are examples of hydrolytic enzymes that serve as important VFs by facilitating fungal invasion of host tissues. Even in the absence of active fungal growth, molds can release toxic secondary metabolites that contribute to tissue damage and disease progression.20 The production of peptide toxins by certain fungi, such as C. albicans, has also been documented, highlighting their potential role in pathogenesis.21 These toxins can be considered classical VFs because, similar to bacterial toxins, they can damage host tissues even in the absence of actively growing pathogens. In addition to toxins, effector proteins represent another important class of VFs in pathogenic microorganisms. These proteins manipulate host cellular functions to benefit the pathogen.22 By interacting with host molecules, effector proteins can alter intracellular signaling pathways, ultimately promoting pathogen survival and proliferation.

Effector proteins are particularly notable for their roles in manipulating host signaling and metabolism, a feature well studied in plant-pathogenic fungi such as Ustilago maydis.22,23 However, the mechanisms and targets of effector proteins in human fungal pathogens remain less well characterized. This knowledge gap limits our understanding of fungal pathogenesis and may hinder the development of targeted antifungal therapies.

While significant progress has been made in understanding bacterial and viral VFs, knowledge of fungal VFs, particularly at the molecular and protein levels, remains comparatively limited. This knowledge gap can hamper the development of effective antifungal treatments and diagnostic tools for infections caused by Candida species and other human fungal pathogens. Computational tools for predicting fungal VFs are also limited. Despite the availability of established methods for bacterial and viral pathogens, including BLAST and machine learning (ML) models such as support vector machines (SVMs) and neural networks,16,24 computational approaches for predicting fungal VFs remain relatively limited. A promising approach to addressing this challenge is the application of ML algorithms. Sachdeva et al. developed SPAAN, a neural-network-based method for predicting adhesins and adhesin-like proteins, which achieved high sensitivity in adhesin prediction.25 However, its application was focused primarily on adhesins. The inherent complexity and diversity of fungal genomes and proteomes present additional challenges for developing reliable predictive models, highlighting the need for specialized bioinformatics approaches.

Existing VF prediction tools, such as VirulentPred,16 HyperVR,26 and EffectorP,27 have demonstrated utility for predicting virulence-associated proteins or effector proteins in bacterial and plant-pathogenic systems. However, tools specifically designed to predict a broad range of VFs in human fungal pathogens remain limited. Although FungalRV28 was developed to predict adhesins in human-pathogenic fungi, its scope is restricted primarily to adhesins, and its application is focused largely on vaccine research. By integrating computational techniques with immunoinformatics, FungalRV facilitates the identification of potential fungal vaccine targets.

Notably, several computational approaches developed for fungal pathogens have been designed primarily for plant-fungus interaction systems. Considerable progress has been made in developing computational methods for identifying fungal effector proteins in plant-pathogenic systems. One example is EffectorP, which predicts secreted effector proteins involved in modulating plant immune responses during infection.27,29 Phytopathogenic fungi produce predominantly small secreted effector proteins; therefore, many of these proteins are less than 200 amino acids in length and contain a high proportion of cysteine residues, which contribute to structural stability and interactions with host plants.29,30 While these identification tools can be useful for studying plant-pathogenic fungi, effector proteins represent only a subset of the broader repertoire of fungal VFs.

The VFs produced by human fungal pathogens comprise an exceptionally diverse collection of proteins that enable pathogens to colonize their hosts, invade human tissues, undergo morphological transitions, and evade host immune responses. Furthermore, VFs produced by human fungal pathogens encompass not only small secreted proteins but also large adhesins, pathogenic enzymes, and regulatory proteins involved in diverse aspects of pathogenesis. For example, the C. albicans virulence-associated proteins ALS3 and HWP2 play important roles in establishing contact between C. albicans and host epithelial cells and in biofilm formation, while BUD4 and VPS11 are examples of proteins involved in morphogenesis and pathogenicity. While computational approaches for predicting effectors associated with plant-fungus interactions have been successful,29,30,31 these methods may not adequately capture the broader repertoire of VFs in human fungal pathogens. EffectorP and Fungtion, for example, are primarily designed to identify secreted effector proteins associated with plant-pathogen interactions.27,31 Consequently, their predictive models rely on features associated with plant-pathogen effector biology, including secretion signals and sequence characteristics typical of small effector proteins. In contrast, virulence determinants in human fungal pathogens encompass a more diverse range of proteins involved in adherence, host tissue invasion, metabolic adaptation to the host environment, and immune evasion. Due to the biological differences between these systems, plant effector prediction tools may not adequately capture the diversity of virulence-associated proteins involved in human fungal infections. These considerations underscore the need for computational tools specifically designed to identify virulence determinants associated with human fungal pathogenesis.

Given these challenges, there is a need for ML frameworks tailored specifically to fungal VF prediction. Such tools can aid in prioritizing candidate virulence markers and potential therapeutic targets, ultimately contributing to improved understanding and management of fungal diseases. Predicted VFs can subsequently be evaluated through domain- and function-based analyses, and their roles interpreted in the context of existing literature to assess their biological relevance.

In this study, we present FunVFPred, an ML-based tool developed to predict VFs in human-pathogenic fungi. While existing tools such as EffectorP and FungalRV are tailored primarily to plant-pathogenic fungal effectors or specific classes of fungal proteins,27,28 FunVFPred aims to provide broader prediction of virulence-associated proteins relevant to human fungal infections. Because of the inherent biological differences between plant and human fungal pathogens, direct benchmarking of VFs in human fungal pathogens against tools trained primarily to predict plant-pathogen effectors may not provide a biologically equivalent comparison. For example, the datasets used to develop EffectorP and Fungtion primarily consist of secreted proteins associated with plant-pathogenic fungi.27,31 Consequently, these tools may not adequately represent the functional diversity and sequence complexity of VFs produced by human fungal pathogens. To address this challenge, FunVFPred utilizes a broad feature set derived from protein sequences to predict virulence-associated proteins across diverse functional classes. To achieve this, we developed a random forest (RF)-based approach trained on protein sequences from Candida species, integrating traditional sequence-based features—amino acid composition (AAC) and dipeptide deviation from expected (DDE) mean—with UniRep embeddings (Figure 1). UniRep, a deep representation learning model pre-trained on millions of protein sequences, captures rich sequence representations that encode structural and functional information in a 1900-dimensional embedding.32

Figure 1: Schematic representation of the workflow of the proposed study(A–I) Pre-processing is shown in (A)–(E). (A) Data collection including positive (virulence factors) and negative data (non-virulence factors); (B) redundancy removal step; (C) the process of balancing the imbalanced data; (D) extraction of conventional features and sequence-based embeddings; (E) the feature fusion step; (F) the data splitting; (G) various classifiers to build model for binary classification; (H) evaluation of classifiers performance to identify the best predictive model; and (I) tool deployment for end-user prediction.

We benchmarked the performance of the RF model against other ML and deep learning (DL) classifiers, including artificial neural networks (ANNs), multi-layer perceptrons (MLPs), and deep neural networks (DNNs). Among the evaluated models, the RF model consistently outperformed the other classifiers in distinguishing virulent from non-virulent fungal proteins. To evaluate its generalizability, FunVFPred was further validated on an independent test set containing sequences from Aspergillus fumigatus. By addressing the need for computational approaches to predict virulence-associated proteins in human fungal pathogens, FunVFPred provides a resource for fungal bioinformatics and may support the prioritization of candidate proteins for subsequent investigation, including potential antifungal drug and vaccine target discovery.