Basically, you can't.

You can try and approximate the line of best fit by drawing a straight line that seems to fit well, but it isn't necessarily the mathematically optimum line. You'll also find that if you do it again, the line will be slightly different. As such, it's not a very good method.

The only way to get a mathematically consistent line is to get the actual values of each data point and plug them into the formula. This also lets you get the R^2 value, which gives you an idea as to how good the fit is.

The line of best fit should start at 0 on your graph, and, after you've drawn it, there should be an equal number of points on each side, or at least roughly an equal number of points.