This is a nice Bayes learning opportunity. It's reasonable to infer that a female-looking username makes someone more likely to be female, maybe twice as likely (not much more than that; this is the internet and people give themselves weird usernames all the time, and actual women may avoid using female-looking usernames in male-dominated forums to avoid drawing attention to their gender). However, the base rate of transsexualism, even within a community as unusual as LW, is still incredibly low and requires a lot of evidence to overcome (e.g. someone telling you they're transsexual).