In light of the current US supreme court case, I'm curious if enough information can be teased out of a DNA sample to get a "reasonable" approximation of the suspect (never mind the legality). I realize the term reasonable is subjective, so characteristics, such as skin color, hair color/texture, approx height, eye color, freckles, etc. I came across this article a few years ago, but the company that made the test went bust in 2009. It could detect race (how accurately?). I seem to recall reading an article a few years ago stating that "soon" we'd be able to get an approximation of appearance from a DNA sample that was better than a sketch artist could provide, though I can't recall now where I read it.
In theory it's possible to have an approximation, but not to know with certainty. Identical twins have the same genomes and look very much alike. Whether it can be done in practice depends on how well we can model the relationship between genes and looks and on how much information is necessary for a judge to permit arresting and questioning a suspect.
Scientists know relatively little about how genes influence physical appearance yet. The height of the suspect is controlled by hundreds of genes with complicated interplay between them, so I don't think it's possible to estimate it just yet. Eye color is more simple: it's controlled by only 3 genes. We know weight is partially controlled by genetics but exactly which genes are responsible is not clear. Age is impossible to tell by DNA alone. Skin color should be predictable, since it's caused by melanin production, and we know which genes are responsible.
So to sum up, DNA analysis can tell the gender, skin color, and eye color of the suspect, but very little about height and weight, and virtually nothing about age. That's not enough information to identify a suspect, although perhaps someone with legal knowledge could comment on this.
But let's say in the future we know enough to determine all these traits. Then DNA analysis would be helpful but never enough to identify a suspect. The problem is that different cells in your body undergo different mutations in their DNA with time. That's how one gets cancer. Let's say there's a mutation in the eye color gene in skill cells from a murderer found on a victim's body. It won't affect the murderer's eye color but it will affect the forensic analysis. So it's impossible to know for sure what the suspect's appearance will be.
As far as I know there are ongoing efforts to find genes that affect forensically relevant traits e.g. facial characteristics and fingerprint pattern type/ridge count, and it's definitely a topic of interest to some law enforcement agencies. However as previously mentioned, there are many factors that could influence complex traits, including in utero environments and other, postnatal effects, such as nutrition, as well as the interaction between those environments and genetic influences. Given past literature showing jury-eligible members tend to over-estimate the importance and statistical accuracy of DNA profiling, however, I think even with future technological and methodological advances in genetic research, this type of evidence should be used with caution, explained with extreme care and thoroughness, and never taken as the single determining factor in reaching a verdict.