Your original selections are your best attempt at maximizing expected value. With no new information, you would be foolish to change it.
However, if you are way behind after Thursday, you have new information which may indicate that your lineup with the highest expected value is unlikely to outscore your opponent's new expected value. In other words, if this weekend's games were played 100 times, your current lineup might average a higher score than any other lineup you could start, but still lose to your opponent most of the time. In this situation, an alternative lineup with higher variance may give you a much better chance of outscoring your opponent (even though it might score fewer points than your previous starting lineup on average). When behind big, you should also try to select players whose performance is uncorrelated with your opponent's players.
So, what are the characteristics of a higher variance line-up?
1. Boom or bust players
2. Line-ups with QB-WR pairs from the same team (e.g. starting Roethlisberger and Holmes in the hopes of a few long Holmes TDs)
3. Line-ups with RBs from different teams than your starting QBs and WRs (there tends to be a ceiling on how many TDs any one team will score in a week--starting multiple players from one team therefore reduces upside relative to starting comparable players from multiple teams)
Player correlation:
If you are up big, starting players whose performance is correlated with your opponent's players is a good move. For example, you might start the QB from the team that your opponents top WR plays for. If both the QB and WR perform poorly, you maintain your lead. If the WR performs great, you are less likely to lose your lead since your QB is also therefore doing well.
If you are down big, the opposite holds true. If you have two WRs with relatively equal projections for the week (say Cotchery and Holmes) and your opponent is starting Roethlisberger, you are better off playing Cotchery.
Final note:
I think these strategies are helpful on the margin, but should never be used over common sense. For example, I would not suggest benching T.O. in favor of Eddie Kennison simply because you are down big and your opponent is starting Romo. These decisions only hold when the players in question have relatively equal projections for the coming week.