Protein-ligand docking is an important and effective structure-based drug design method widely used for large-scale screening of drug candidates. The core of a protein-ligand docking program consists of a sampling algorithm and a scoring function, which produce different poses of a ligand and estimate a score for the pose with respect to how good it reproduces the native conformation of the ligand at the protein binding site, respectively. Nature-inspired algorithms such as particle swarm optimization (PSO) and firefly algorithm (FA) are emerging optimization techniques for simulating social behavior of creatures and the nature-based law of the survival of the fittest. In this study, we investigated the application of FA in ligand pose prediction using a protein-ligand docking program PSOVina. Importantly, we tested four strategies on the classical FA to enhance the protein-ligand docking performance, namely application of the logistic map to diversify the search, Mantegna's method for simulating Lévy flight to generate random walk, elite selection to inherit better solutions across iteration, and k-mean clustering to find more than one optimum solutions. We performed parametric analysis of FA and benchmark tests using two datasets, PDBbind database (version 2014) and Astex Diverse set. The results show that although the average relative mean standard deviation (RMSD) of predicted poses is not always the best, our FA variants are better than those of PSOVina in average success rate, suggesting that FA has potential usefulness for performing robust searches in the ligand conformational space.