If DSATUR heuristic was proven efficient for many instances of the vertex coloring problem, this paper aims to analyze inefficiency causes of DSATUR heuristic, extending DSATUR with matheuristic operators. Using an Integer Linear Programming formulation allows to have larger local greedy optimization in DSATUR construction scheme. Matheuristics allow also to initialize DSATUR heuristics and matheuristics with cliques or a partial optimal coloring. Dual bounds are also obtained, improving lower bounds implied by cliques computed by matheuristics. Computational analyses are provided, highlighting the strengths and weaknesses of DATUR heuristic and matheuristics.