Toward Comprehensive Benchmarking of the Biological Knowledge of Frontier Large Language Models