Tackling the widespread and critical impact of batch effects in high-throughput data