New drugs are usually tested and perfected in mice, before they are ever given to humans. Mice make dependable subjects in a lab, and a drug that proves safe and effective in a mouse may very well be safe for humans.
But, alas, a mouse is not a man. Too often, drugs show an effect in animals, but have little effect in humans. More importantly, drugs that prove safe for animals may have unwanted side-effects in humans.
For this reason, scientists are trying to make mice more "human-like", to make their drug response more predictive of human drug response. Mice have different genes from humans (although there is at least a 95% DNA similarity), and this leads them to metabolize drugs differently. So mice with more human-like genes are being genetically engineered for use in drug-testing. A mouse can be more like a man, after all.
Yet any study (whether mouse or human) adds to the cost of developing new drugs. So some scientists are also attempting to develop predictive computer models (in silico), to simulate (on a computer) the drug's effect on the body (such as "predicted liver toxicity"), without requiring as many animal studies.
But human studies will always be needed to test new drugs. The human body is too complex and our genes are too unique from other animals to fully rely on other predictive models.