An adaptation of particle swarm clustering applied in basal cell carcinoma, squamous cell carcinoma of the skin and actinic keratosis Fabiano de Oliveira Poswar a , Laércio Ives Santos d , Lucyana Conceição Farias a , Talita Antunes Guimarães a , Sérgio Henrique Souza Santos b , Kimberly Marie Jones c , Alfredo Maurício Batista de Paula a , Reinaldo Martinez Palhares e , Marcos Flávio Silveira Vasconcelos D'Angelo f, ⁎, André Luiz Sena Guimarães a, ⁎ a Department of Dentistry, State Universit of Montes Claros, Brazil b Institute of Agricultural Sciences, Food Engineering College, Universidade Federal de Minas Gerais (UFMG), Montes Claros, Minas Gerais, Brazil c Center for Research, The Educative Association of Brazil, Montes Claros, Brazil d Campus Montes Claros, Federal Institute of Minas Gerais, Brazil e Department of Electronics Engineering, Federal University of Minas Gerais, Brazil f Department of Computer Science, State Universit of Montes Claros, Brazil abstract article info Article history: Received 19 October 2016 Revised 16 December 2016 Accepted 20 January 2017 Available online 24 January 2017 Introduction: This study used the comparison of basal cell carcinoma (BCC), squamous cell carcinoma of the skin (SCC) and actinic keratosis (AK) to test a new method for data set clustering in the leader gene approach. Methods: Genes related to BCC, SCC and AK, were identified in the databases: OMIM, Genecards and NCBI Gene. A network was built for BCC, SCC and AK using STRING. For each gene, a weighted number of links (WNL) was cal- culated based on the combined STRING scores. The genes were then clustered according to their WNL and TIS, using an adaptation of particle swarm clustering (PSC) or K-means clustering. Results: A disagreement between K-means clustering and PSC was observed for both BCC and SCC. PSC suggested completed different leader genes to BCC and SCC. While K-means clustering indicated that CTNNB1 and TP53 were associated with BCC and SCC. In contrast, no differences in methods were observed to AK, which had the shorter network. TP53 was the only leader gene for AK. Conclusion: In conclusion, the current study suggests that PSC is an interesting tool for clustering genes in bioin- formatics analyses of prevalent diseases. K-means clustering should be used in the small network. The current study also suggests TP53 may play a central role for AK. Additionally, CTNNB1 seems to be related to BCC, while CTNNA1 is related to SCC © 2017 Elsevier B.V. All rights reserved. Keywords: Particle swarm clustering Skin cancers Potential malignant lesion Metastasis Bioinformatics 1. Introduction Non-melanoma skin cancer, including basal cell carcinoma (BCC) and squamous cell carcinoma of the skin (SCC), are the most common malignancies in humans (Guenther et al., 2015). BCC and SCC share es- sential characteristics. They both are derived from epidermal keratinocytes and are associated with ultraviolet light exposure, fair skin, and immunosuppression (Madan et al., 2010). Nevertheless, SCC has a greater tendency to metastasize and frequently arises from pre- cursor lesions, such as actinic keratosis (AK) (Chetty et al., 2015), while BCC only very rarely metastasize and it arises directly from healthy skin (Bauer et al., 2011). AK is a lesion related to cumulative sun exposure. After AK is established, it can either evolve to spontane- ous remission, remain stable or transform into invasive SCC (Berman and Cockerell, 2013). The identification of biological markers that could be related to the distinct behaviors of BCC and SCC is an exciting field of research since it has the potential to unveil key elements related to invasion and me- tastasis (Shimizu et al., 2001; Galer et al., 2011; Yin et al., 2013). Also, the discovery of genes differentially expressed between AK and SCC can be important in understanding the carcinogenesis of keratinocyte- derived neoplasms. In fact, much research has been done comparing these neoplasms, and there is already a good amount of data on this topic (Poswar et al., 2013; de Oliveira Poswar et al., 2015). Bioinformat- ics has emerged as an important tool for mining relevant databases for significant evidence of specific pathways and, in particular, in identify- ing genes that exhibit a high level of activity with other genes during le- sion-specific disease processes (Poswar Fde et al., 2015). The identification and comparison of disease-specific “leader” genes have been identified as an approach with promising potential for Meta Gene 12 (2017) 72–77 ⁎ Corresponding authors at: Universidade Estadual de Montes Claros, Hospital Universitário Clemente Faria, Laboratório de Pesquisa em Saúde, 562 Cula Mangabeira Avenue, Santo Expedito, 39401-001 Montes Claros, MG, Brazil. E-mail addresses: marcos.dangelo@unimontes.br (M.F.S.V. D'Angelo), andreluizguimaraes@gmail.com (A.L.S. Guimarães). http://dx.doi.org/10.1016/j.mgene.2017.01.007 2214-5400/© 2017 Elsevier B.V. All rights reserved. Contents lists available at ScienceDirect Meta Gene journal homepage: www.elsevier.com/locate/mgene