You're right. And thats the point. Circumstantial evidence including DNA evidence should never be sufficient to convince us that there is evidence beyond a reasonable doubt. You should need an eyewitness, clear video footage, etc in addition to a DNA match, etc. Simply because someone happens to be at a location at a given time or around that time and made calls to the person murdered, etc. might make them a suspect, but it shouldn't be enough to convict. Nor should DNA evidence alone. With a mere 99.8% accuracy that means that with billions of people on the planet that without additional evidence there will be wrongful convictions if used alone to gain a conviction.
First off you are crossing back and forth between not liking circumstantial evidence and not liking the accuracy of DNA evidence. Even if DNA evidence was 100% effective, you still have the issue of circumstantial evidence tying them to the actual murder (instead of just to the knife). And if you have eyewitness or clear video footage you may not have to rely on circumstantial evidence, but you do have to deal with the same accuracy problems of DNA. Even the most high resolution video camera probably has far less reliability than DNA. Perhaps they just look like the murderer, just like their DNA could look like the murderer's DNA.
The problem is not circumstantial evidence or inaccurate evidence. The problem is not enough evidence. Placing the suspect's DNA at the scene may only make you 99.8% guilty, which potentially means about 14 million people on Earth look as guilty as you. But placing you in the same neighborhood during the weekend the murder took place may bring that number down to 20 people. I would be fine with the police questioning all 20 of those people if they couldn't narrow the search more.
People need to understand that no one has ever been convicted with 100% certainty in the history of history. Perhaps with 99.999999999999999% certainty, but not 100%. Even evidence that is only 2% accurate is useful when combined with a large amount of other evidence (like being in the same neighborhood as the victim, which is probably closer to 0.01% accurate by itself).
As our society's data crunching abilities grow, we may start to create more quantitative estimates of guilt. We could set a threshold of guilt that has an acceptable amount of false convictions. Say 1 false conviction in 10,000 accurate convictions. Regardless of any delusional ideas that our justice system is set up to never convict innocent people, we will always need to leave room for doubt. That is why we say "beyond a reasonable doubt" instead of "beyond any doubt".