The Mycobacterium tuberculosis (Mtb) genome encodes hundreds of proteins with unknown or putative function based on sequence similarity to characterised proteins. Incomplete knowledge of Mtb functional genomics undermines efforts to understand the complex biological repertoire of this deadly human pathogen. While bioinformatics provides useful clues about gene function, biochemical characterisation of gene products remains essential to accurate annotation, particularly in cases where functional redundancy is suggested based on the presence of multiple homologous proteins. The Mtb genome includes 7 genes which encode conserved hypothetical proteins with unknown function that are annotated as “pyridoxine 5’-phosphate (PNPOx)-like” proteins based on structural similarity: Rv2607, Rv2991, Rv1155, Rv2074, Rv3369c, Rv1875, Rv0121c. PNPOx is a flavin mononucleotide (FMN)-dependent enzyme involved in the biosynthesis of the crucial cofactor, pyridoxal 5’- phosphate (PLP). The work presented in this thesis shows that while Rv2607 exhibits canonical PNPOx activity, Rv1155 and Rv2991 do not have affinity for FMN as do other members of the PNPOx class. They instead bind tightly to the unusual flavin coenzyme F420 by isothermal calorimetry (ITC), saturation transfer difference (STD) NMR, and X-ray crystallography. Although recent studies have drawn attention to F420 due to its role in activating the promising anti-tuberculosis drug PA-824, little is known about the function of this coenzyme in Mtb metabolism. In order to identify the substrate recognition properties of Rv1155 and Rv2991 respectively, a novel application of fragment-based approaches was employed. A library of fragments was designed based on known ligands to flavoenzymes and screened for binding to Rv1155 and Rv2991 in the presence and absence of F420 using STD NMR and differential scanning fluorimetry. Rv1155 and Rv2991 show affinity for distinct classes of fragments. Using a computational method known as “virtual fragment linking,” fragment binding data was used to predict and test potential substrates. This approach to probing structure-function relationships may serve as a generalisable method for exploring functional diversity within a structural class of proteins.